September 12, 2025 • 62 mins
What happens when advanced mathematics meets manufacturing? The result is a new way of creating products that range from record-breaking running shoes to life-changing medical devices.
In this episode, we sit down with Elissa Ross, mathematician and CEO of Metafold 3D, to explore how her company is using mathematics to reshape design and manufacturing. Metafold’s platform is built on geometric intelligence which is her company's novel approach that transforms shapes into data that can be analyzed and improved. Instead of simply representing objects the way traditional CAD software does this method allows engineers to simulate and optimize designs with greater speed and accuracy.
At the core of this approach is implicit modeling with signed distance functions. While it may sound highly abstract, it has very practical applications. This technique allows manufacturers can run thousands of simulations in the time it would normally take to analyze a single design. The result is shorter development cycles and products that perform better in the real world.
Ross also reflects on her journey building Metafold, from its early focus on 3D printing to its current role serving major industries. She explains how their API-first platform gives customers the flexibility to solve specific challenges, such as analyzing tolerances, reusing similar parts, or predicting how designs will hold up under different conditions.
Whether you are a designer, an engineer, or simply curious about how mathematics is shaping the technologies around us, this episode offers a thoughtful look at the future of manufacturing intelligence.
Shownotes: Check back soon
Podcast Engineer: Faith Fernandes
Subscribe to our premium version and support the show.
Follow us:
Twitter
Instagram
Linkedin
3DHEALS Website
Facebook
Facebook Group
Youtube channel
About Pitch3D
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Hello there, welcome to the Lattice podcast, episode
number 93.
Today, we explore the worldwhere mathematics meets
manufacturing innovation withElisa Ross, co-founder and CEO
of Toronto-based startup Metafo3D.
Elisa and her team aretransforming the way industries
analyze and understand 3Dgeometry, using advanced
(00:24):
geometric intelligence andimplicit modeling.
Whether you are a designer,engineer or anyone fascinated by
the creative side of math, getready for an eye-opening
conversation about shape and thefuture of making things.
Please listen to the disclaimerat the end of this podcast.
Please listen to the disclaimerat the end of this podcast.
(00:47):
Hello, hello.
Thanks for joining the pod.
Today we have the CEO andco-founder of Metaphone 3D,
elisa Ross.
Hello Well, elisa.
I feel like today's interviewis a bit like a challenge for me
because, even though I lovemath, this is like one of those
mathematical challenge Iencountered when I was in school
(01:08):
.
It's both exciting and kind offascinating, but also I
understand the limitation of myknowledge.
So first of all, I'd love tojust hear a little bit about
what Metafold 3D is and a littlebit about yourself what
Metafold 3D is and a little bitabout yourself.
Speaker 2 (01:30):
Yeah, absolutely Happy to Well.
First of all, thank you, jenny,for the invitation.
I'm super excited to be hereand I promise we won't go too
deep on any mathematical topics.
Metafold 3D is a startup.
We're about five years old now,we're based in Toronto and
we're all about math.
As you've already introduced,I'm a mathematician by training,
specifically geometry, and wecan get into that if you like.
(01:52):
But Metafold is all aboutbringing geometric insight to
manufacturing, and our productis called Geometric Intelligence
, and it's really all aboutturning geometry into
information that can be used tomake decisions.
We can unpack all of that, butthat's the highest level.
Speaker 1 (02:16):
Awesome, I love it.
Yeah, I think we want to unpacksome of the maybe the most
important concepts of yoursoftware, because I am also
learning as we are doing thisinterview, and why don't you
unpack some of the mostimportant concepts at the core
of your software?
Speaker 2 (02:37):
Sure.
So geometric intelligence isour technology and it's really a
platform for shape intelligence, to shape information and
analysis.
So what I mean by that is thatwe take geometry as input and we
give people using our softwareinformation about that geometry.
(02:57):
So why would you want this?
The idea is really that youknow, shapes are not that
meaningful on their own when wemanufacture things.
These shapes exist in the worldto perform a purpose, to serve
a purpose, and so our approachis all about helping people
(03:20):
understand how their shapeswould perform or how they will
be manufactured in the worldbefore they go ahead and make
those things.
So, very specifically, weconvert geometry into
information about performance.
Like you know, how well will itperform under certain loading
(03:41):
conditions?
Let's say, we convert it intomanufacturing information.
We also do a similarity.
We can look at similarities, soanswer questions like have I
made something like this in thepast?
And we can also address thingslike tolerances, which have a
huge impact on manufacturing atscale.
So, again, we can get into allof those topics, but the thing I
(04:04):
really want to focus on oremphasize is that we convert
geometry into information thatcan be used to make decisions
Under the hood, of course, andwe can get into the mathematics
of it under the hood.
We take a fairly differentapproach from traditional CAD
software.
Speaker 1 (04:22):
Yeah, I want to unpack the concept of CAD before
because I had the samemisconception about it, since I
am not a designer and I don'twork with it day in, day out.
But CAD software itself doesinclude analytics and also
provides some kind of analysisfor manufacturing and outcome.
(04:44):
I mean, some of the biggercompanies that I have in mind
are Autodesk or SynopsysSynopsys sorry they are.
They also have designcapability and analytics
capability.
Where do you guys are you?
Are you guys part of this chainof information or you are
(05:07):
completely different from that?
Speaker 2 (05:10):
Yeah, you're right, all of these platforms do
provide some kind of analysiscapability, and the reason is
what I stated before, that theshape is kind of useless on its
own.
You need to know something aboutthe shape.
Where we really differ fromthose platforms is in the way we
represent geometry, and thatgoes into the details of our
technology.
But, in short, we representgeometry using an implicit
(05:35):
approach which is different fromthe way traditional CAD is
represented, and this allows usto extract different kinds of
information and do thingsdifferently.
So it means that we have, forinstance, a faster path to
simulation, and so if you'redoing simulations one at a time,
(05:58):
then conventional CAD isprobably a great choice.
If you're trying to do athousand simulations at once,
then maybe we should considerhaving a conversation.
This is the kind of thing wecan help with.
So that's actually.
Another component of ourtechnology is that we are really
built around working with largevolumes of 3D data.
(06:18):
So if you have a lot ofgeometry, you're trying to do a
lot of experimentation or you'regoing to produce a lot of parts
, then this is a very good spotfor our technology to slot in,
because if you use conventionalCAD, you're probably looking at
those things one at a time,which is very time-consuming.
Speaker 1 (06:38):
And yeah, I have to say this is a very enlightening
conversation for me personally,because decades ago, believe it
or not, I actually did researchin simulation of a 3D shape
which is aneurysm, because wewant to use a mesh to predict
what kind of aneurysm shapewould rupture.
And actually that kind ofinformation now translated to a
(06:59):
company called the HeartFlow anda bunch of other
medical-related simulation toolsto predict heart attacks and
aneurysm, ru related simulationtools to predict, you know,
heart attacks and aneurysmrupturing medical device, but
they're all based on thetraditional CAD.
Now I thinking, thinking back,and I remember I have to wait
for a whole day for the analysisto come in, for the stress
(07:19):
sheer stress calculation and,yes, just to create the mesh
itself takes a couple hours andwe're talking about circa 2010.
And I remember that I work withPhD, so like I will like figure
out the shape of the aneurysmto be simulated, and then he
would keep the computer onovernight to generate something
(07:41):
the next day.
It's either a fail or success.
It's tremendous amount of work,but the results you know.
It just takes so long.
I remember that had we hadimplicit modeling capability,
things would be different andmaybe we can test 10,000
different aneurysm models.
Speaker 2 (08:00):
This is possible.
Yeah, I mean it's also possiblethat that approach was actually
great and that it was just fineIn some cases, depending on the
type of simulation you'rerunning, and then they need to
change the format of the datainto a different format for
(08:27):
simulation.
Maybe it's a different person,that's what you described.
There's someone else whoactually runs the simulation and
so, you know, in largecompanies this might be a fully
separate team.
So if you're trying to iterateon a design and you're doing
that super slow process, thisjust takes too much time.
What we can offer is helpingpeople bring a lot more of that
(08:48):
information to the designerreally actually in some cases so
that they can run theirsimulations and get information
that can inform their designprocess much more immediately.
Speaker 1 (09:00):
So are there certain type of shapes in particularly
would benefit from Metafo'ssoftware.
Speaker 2 (09:09):
I wouldn't say there are certain types of shapes.
No, I think what's kind ofinteresting to understand is
that so much of the way weinteract with geometry on a
computer will come down to howyou represent geometry.
And so for us it's not so muchthat there's certain shapes that
are more amenable to approachthan others, it's more kind of
(09:33):
levels of accuracy, let's put itlike this, and depending on
what you're trying to achieve,you might choose different
representations.
So for manufacturing, youprobably need a traditional
representation like a B-rep oreven like a test-related format,
like the STL probably familiarto you in the 3D printing world.
(09:54):
But for what?
Our perspective at Modifold isthat if you're trying to get
information about the shape, itcan be very helpful to move to
an implicit representationbecause you have all this
information about the shape.
It can be very helpful to moveto an implicit representation
because you have all thisinformation about the volume of
the structure and you can dooperations on it that give you
information that you can't getwith a traditional format.
(10:15):
So it's not so much about thetype of shape as it is maybe in
the kind of information you'retrying to extract from your
geometry.
Speaker 1 (10:23):
Alyssa, I view you as my mathematic teacher, so there
are two concepts I'm hopingthat we can unpack for average
people who listen to our podcast.
One is implicit modeling, whichis a new concept to me, and the
other one is sign distancefunction.
I know they sound like bigwords, but actually in reality,
after a while, I feel likethey're actually not that
(10:45):
complicated to understand.
Speaker 2 (10:47):
Yes, I agree, and actually they're they're really
one in the same in a lot of ways.
So let's start with the signeddistance function.
This is a very um, simplemathematical function and it it
operates on three-dimensionalspace.
So you begin with a shape, asurface in three-dimensional
(11:07):
shape space rather, and the sinedistance function is defined by
that shape and for every pointin three-dimensional space it
returns the distance from thatpoint to the surface of the
shape.
So this is a function that iszero, I would say.
In mathematical speaking.
(11:28):
We would say identically zeroat the surface of the shape.
Everywhere else it is non-zeroand is actually equal to the
distance from the shape.
If you're outside the shape,it's positive, if you're inside,
the convention is that it isnegative.
So already this gives us such arich space of information that's
(11:49):
not present in, let's say, atriangle mesh.
So for comparison, a trianglemesh is a set of triangles that
you stitch together toapproximate the surface of a
geometry.
Important to understand in bothcases.
This is not geometry, it's arepresentation of geometry, and
(12:11):
I like to make that pointbecause people often have kind
of a mental idea of whatgeometry is, what it looks like,
and that is always approximatedon a computer.
So the sine distance functionis a precise and complete
definition of the shape.
But when we represent it on acomputer we probably need to
(12:32):
sample a grid to, like you know,sample the distance across the
grid of values to reallyunderstand where that shape is.
So that's the concept of signdistance function, Clear so far.
Speaker 1 (12:46):
Yes, so the mesh and the sign distance function.
Would it be accurate to saythey're both kind of ways,
different ways to represent a 3Dgeometry?
Speaker 2 (12:59):
Absolutely, absolutely.
And there's a third way.
There's multiple ways, but athird common way is the boundary
representation format, the BREP, and this builds up surfaces
through points, lines andsubsurfaces basically too
(13:20):
listeners.
Speaker 1 (13:22):
Well, so I want to give us some real-life examples,
if you're allowed to, or maybesome hypothetical examples, of
how Metafold works, as opposedto, let's say, autocad and all
the other CAD software that hassome kind of intrinsic analytic
(13:43):
capability, and also for peoplewho are already working with
those kind of softwares can theyuse your program seamlessly,
kind of integrated?
Speaker 2 (13:49):
into it.
Yeah, these are.
These are good questions, um,maybe just because I want to
close the loop on this.
You did ask about implicitmodeling.
Yeah, I want to say thatimplicit modeling is just refers
to using signed distance fields, usually to build up more
complex geometric shapes.
So, like conventional modeling,you're usually, you know, in a
(14:12):
boundary representation orcreating a triangle mesh, and
when you're doing implicitmodeling, you're using some kind
of volumetric representation,probably a signed distance field
, and this is actually kind of abeautiful thing.
If you think about signeddistance fields, you can
represent them as a grid ofvalues.
So if you, if you want to say,add two together, it's as simple
(14:33):
as arithmetic on that grid.
So certain operations becomemuch simpler in implicit
modeling world, which is whypeople tend to like it.
So, from uh, from that to yourquestion about real world
applications.
Yeah, so I'd love to give someapplications and I, given your
(14:54):
audiences is in the health space, I'll try and keep them kind of
health care adjacent.
Um, and actually, on that, oneof the the big application areas
for us is in performancesportswear, specifically
footwear, so running shoes, andhere the goal is to create the
(15:15):
most performant running shoespossible.
So we work with teams who areworking on that problem, that
problem.
There's been many materialdevelopments over the past
couple of years and also justdevelopments in the way the
materials and geometry are puttogether that have created
pretty remarkable outcomes interms of athletes, like we've
(15:37):
seen the marathon records beingbroken and all sorts of other
similar records being broken,largely by footwear technology.
So there's a real interest inpushing forward this technology.
The standard process forproducing footwear is to come up
with an idea, physicallyfabricate it and test it.
(15:58):
Okay, so this is certainly agood way of getting information
about geometry and material.
However, it's very slow, soteams will spend months in these
prototyping cycles trying toiterate and find the best
performing shoe, but in fact,it's not an exhaustive search
(16:19):
because that is just such atime-consuming process.
So what we help with here istaking a broad sweep of geometry
, taking a material palettethat's potentially specific to
that manufacturer, and thenhelping them scale up their
(16:40):
experiments.
So really simply, all we do isreplicate their physical tests
digitally and then radicallyscale them up so that they can
simply get so much moreinformation than they would
through a physical or manualprocess.
That makes total sense.
Yeah, and that is.
(17:02):
You could imagine this sameprocess working in a more
healthcare related field, whereyou're trying to really, you
know, provide some level ofcushioning and you want to make
sure that you're achievingexactly the right response to
force, depending on theapplication.
But often we see the samecombination of geometry and
(17:26):
material.
You want to vary both of themsystematically so that you can
understand how your partperforms.
Speaker 1 (17:34):
Yeah, I mean, I can see this can definitely
translate, not just in the shoeindustry, but all kinds of other
wearables or pathetics.
What about implants?
Would that be?
If people want to iteratedifferent designs for hip
implants, for example, can theyuse your software to somehow
scale up their design process?
Speaker 2 (17:57):
Yeah, absolutely.
I mean, it's the same kind ofthing.
What is sort of interestingabout the implant world, though,
as I understand it, being anoutsider to the space is that it
is patient-specific, and sothat element might indicate that
you know you're probably for agiven patient.
You might not want to spend toomuch time, you know, physically
(18:20):
testing, and that mightactually make it a very good use
case for a small digitalexploration, like given the
patient geometry.
We're trying to find the bestperforming implant.
Here's a systematic way ofstudying this.
It takes, you know, an hour orwhatever.
It takes digitally, and then atthe end of it you find the
candidate that works the bestand will yield the best outcomes
(18:43):
.
Speaker 1 (18:44):
Well, as much as 3D printing industry would like to
have personalized implants, thereality is the majority of the
implants are not personalized,but more and more implants are
now 3D printed.
I think even just like onelevel up improvement in design
not personalized, but let's saymodel certain X or YZ can be a
(19:07):
better fitting for osteoporoticpatient and stuff like that Like
just incremental improvement.
They need design iterations aswell.
Speaker 2 (19:18):
Yeah, exactly, and that's a great point, that a lot
of these are based on kind ofbest practices as opposed to
lengthy iteration cycles to findthe optimal solution, or not
necessarily the optimal, but anoptimal solution.
So yeah, absolutely, that wouldbe a great fit.
Speaker 1 (19:36):
And if I'm not?
Speaker 2 (19:36):
this speaks to kind of like a bigger theme for us as
a startup, which is that wehave really moved away from
focusing on 3D printing.
So we, you know, when we began,we away from focusing on 3D
printing.
So when we began, we were veryfocused on 3D printing because
there are all kinds ofinteresting geometric things.
Of course that's how weconnected, but as a market it's
(19:59):
just not quite growing the waythat we believe it should and
hope it will.
But more than that, inconventional manufacturing,
whether you're talking aboutmedical or footwear or aerospace
, there are so many interestinggeometry problems and
interesting places to optimize.
So, yeah, but understandingperformance is one of our key
(20:20):
pillars of turning geometry intoinformation.
It's not the only one, so wealso work in instance or
on-demand manufacturing.
We work in kind of similarityanalysis for part reuse and also
tolerancing.
I don't know if any of thoseresonate for you, Denny, or want
(20:40):
to pursue further.
Speaker 1 (20:41):
I want to go back to the fact that you pivoted away
not away, but like expanded yourhorizon significantly because
you discovered market and wefrequently interview people who
did that, because I think that'sa natural progression of
knowledge and insights as anentrepreneur.
And it's actually good for you,because 3D printing industry is
(21:04):
a small industry and it stillis growing very fast.
But to survive as a company,you need to discover new markets
, you need to create the market,and I'm just surprised in a
good way that you have found amuch bigger market that is not
well served by the currentsoftware providers.
Speaker 2 (21:25):
Yeah, this is.
It's exciting for sure, and youhighlight a bit of a point here
, which is that the currentsoftware providers and CAD in
general is a very old and stickykind of set of technologies of
geometry we were talking aboutearlier.
(21:46):
They exist both because theyare pretty good, but also just
because they exist and they'vegot inertia and they've got
people who've got decades offiles in those formats.
So it is interesting to beapproaching this large market of
manufacturing with a newapproach to shape analysis and
(22:06):
geometry understanding, and wedo certainly believe that
there's a lot to be gained fromthis approach that's distinct
from what the kind ofconventional players are doing.
Speaker 1 (22:18):
Yeah, and also I feel like the market probably is
also slowly responding to youand your company's product
because, like you said, theyweren't aware of new ways of
doing things because of theinertia.
And how do you tell people youknow this way is a better way or
more productive for you?
Speaker 2 (22:38):
I mean, I think it's.
I think that's been kind of akey understanding for us too, is
that it's not really themessage is not hey guys, this is
the better way, it's more.
What more can we do withgeometry?
Ultimately, everyone wants thesame thing.
They are using geometry as atool to produce something, and
(23:03):
the last thing they want to bedoing is thinking about their
geometry representation.
They just want to design athing, they want to understand
how it's going to work andproduce it, and then obviously,
they want to make that veryprofitable.
And so to make it profitable,you ultimately need more
information, and so thisunderstanding from us.
(23:24):
It's not about changing the waypeople design things.
It's about taking the thingsthey're already doing and giving
them more information aboutthem so that they can make
better things, make fasterthings, increase their yield.
You know, stop redesigning thesame part over and over all
these kinds of things.
Speaker 1 (23:44):
I think that you're one of the companies that is
cloud native, the nextgeneration of software provider,
and now you're also going APIfirst strategy in terms of
business development.
And the fact is, it's not justgeometry analysis.
You also provide a suite ofbenefits like instant quoting,
(24:08):
tolerance analysis, like yousaid, and smart part reuse.
I'd love to hear you justunpack some of those features,
because I don't see this veryoften.
Speaker 2 (24:24):
Yeah, for sure, and I think like a couple of things
here.
So, cloud native yes, we have acloud offering.
However, we are not exclusivelycloud and this has been like
another sort of semi-painfullesson to learn right that if
there's certain industries, forexample, aerospace and even, to
some extent, sportswear, wherethey just like they don't want
to be on the cloud, and I get it.
So we have solutions for peoplewho want to achieve dramatic
scale using the cloud and wehave solutions for people who
(24:45):
want to run it on the computerunder their desk.
This is all possible.
But yeah, so in terms of theother offerings we have, it is
all rooted in the sametechnology.
So using signed distancefunctions to gain understanding
about shapes For the InstantCADquotation super interesting
(25:07):
problem.
Quotation Super interestingproblem.
And I think increasingly we'reseeing more and more
manufacturing that people aretrying to manufacture more
domestically for all thegeopolitical reasons.
But really interesting problemscome up in the on-demand
manufacturing space.
So the idea in that business isthat those on-demand
(25:29):
manufacturers they receive apart and they need to provide a
quote as fast as possible.
So they need to provide a quoteto the customer on how much
it'll cost the customer to buy,but they also need to have a
supplier cost.
They need to provide, you know,like put it out for their
suppliers to produce for acertain price.
So how do they get thatinformation?
(25:50):
Certainly, it's a function ofthings like material and lead
time and quantity and a wholebunch of non-geometric factors,
but at some point there's a lotof shape analysis that needs to
come in there.
So, understanding the part,what kind of machine do you need
to make it on and what are thefactors that would make it more
(26:12):
or less complex to produce?
How many hours would it take toproduce this?
So we don't do those things.
What we do is we convert thegeometry into a set of features,
geometric features and otherforms of data that those
companies can then ingest andbuild their costing model.
(26:34):
So I hope that makes sense.
We're not providing the price,we're not providing the process.
We're converting the geometryinto other forms of information,
other measurements, differentkinds of analysis results that
allow them to create models thatare accurate for pricing.
Speaker 1 (26:53):
Does it include manufacturing process materials
and that kind of information inyour geometry outputs?
Speaker 2 (27:01):
So we are really strictly on the geometry side,
so we provide the geometricfeatures.
They bring all thenon-geometric features and
that's why they are building thepricing model themselves.
But we can, you know, via ourAPI, just give them all of this
data about the geometry, andwhat this means is that fewer
(27:22):
parts where an expert needs toopen up the file, just one at a
time, spin it around, estimatethe time it would take to
produce it, which is kind of thestate of things at this point.
There's some automation andthere's some expert quoting that
has to go on.
So how do we increase theamount of automation for that
process?
Speaker 1 (27:44):
I definitely understand that you guys are
purely in the geometry business.
However, doesn't it soundlucrative if you are able to
create an end-to-end pricingquote instantaneously?
Speaker 2 (27:55):
Let me tell you this is quite a hard problem.
Yes, it is definitely lucrative.
And there are businesses thatare operating on this basis, for
sure.
Speaker 1 (28:06):
I think I can imagine why some of the manufacturers
would not want to share the datathat they use because it's
proprietary for their ownprocess.
Definitely, yeah.
And also, rome is not buildingone day.
Metaphode is not coming out ofthin air, so I want to just go
back to your original founderstory, if that's possible, to
(28:28):
tell us a little bit about yourearly days before Metaphone and
how that translates into today.
Speaker 2 (28:35):
Sure, I don't know how far back you want to go.
Speaker 1 (28:39):
By the way, just curious, I was reading the
history of cat, which I willshare a link in a podcast.
Are you related in any way withDoug Ross?
Speaker 2 (28:48):
I am not.
I'm not in any way related tothat person.
Okay.
Speaker 1 (28:52):
We invented the term cat, basically.
Oh, I didn't know that.
Speaker 2 (28:56):
Okay, well, maybe I should claim I'm related and
have more credibility.
Speaker 1 (29:00):
He invented the term computer-aided design in the
1960s and his name is Doug Ross.
Mit guy permanently imprintedhis name in the industry.
Speaker 2 (29:13):
Wow.
Well, I can aspire to that kindof goal.
I mean, there are a lot ofRosses in the world.
At this point I'd definitelyrather that metaphor make it
smart rather than me personally.
But so, yeah, I am not relatedto Doug Ross my background so
I'm a mathematician by training.
But it may be worth saying thatI never set out to be a
(29:36):
mathematician.
This is not what I thought Iwould be doing.
If you had told me this is theway things would work out when I
was a kid, I would be laughing.
I really wanted to be an artistand math was just not on my
radar.
I'm a very, very visual thinkerand so I got into math through
art.
I was really fascinated bytilings and patterns, these
(29:58):
two-dimensional patterns andkind of geometry stuff.
And then that interest carriedme through first an
undergraduate and the master'sand then finally the PhD.
But I've always been veryfascinated by the interaction
between geometry andapplications.
And it's a little unusualbecause I'm more interested in
(30:20):
pure math and applying ideasfrom pure math to problems in
the real world.
This is in contrast to thefield of applied math, which
always begins with a questioncoming from another field of
science usually, and there's awhole set of mathematical
techniques that fall under theapplied math, but that's not
(30:40):
really been my interest.
I've been interested in theoryof geometry and combinatorics
and things like this thatnevertheless have applications
to especially material science.
So my PhD was on, really, thestructure of zeolite materials
that have a very interestingmolecular structure and this has
led to an interest in latticestructures and metamaterials,
(31:03):
which has, you know, inspired alot of my work in 3D printing.
But so all that to saymathematician but kind of like
an uneasy relationship withmathematics in certain ways, and
very interested in art, indesign, and so, following, you
know, the whole academic path, Imet the person who's now my
(31:26):
co-founder.
This is Daniel Hamilton, whowas working.
He had started his ownconsulting business applying
mathematics to architecturebasically, and so, given my
interest, you know, we clickedright away and I started working
for him and then we workedtogether for quite some time and
(31:46):
eventually kind of became likea partnership in that business.
And then eventually, you know,we had a small team and one of
those team members, tomRoslinski, joined the three, the
two of us and the three peoplestarted Metafold.
So yeah, and Metafold I guess Ican say was inspired by the
(32:10):
consulting work we were doing.
So in this consultancy, likewho is it who is hiring
professional mathematicians tosolve problems?
So we mentioned architecture.
But we ended up working, infact, in sportswear for a
company that was trying to dosomething really ambitious using
3D printing.
And it was that context likethey're literally hiring
(32:34):
mathematicians because they haveproblems they can't solve about
3D geometry.
What are the like, what's goingwrong here and what's not
functioning about their existingsoftware for CAD, for designing
and understanding those shapes.
So that was the kind of themotivation for Metafold.
Speaker 1 (32:56):
And now it's been five years and I think I've seen
you guys grew at least threeyears now.
I've known you three years agomaybe.
What are some of the ahamoments for you?
Because I know you're very opento new ideas and ready to pivot
whenever, um, it's necessary.
Like, what are some of the ahamoments for you throughout this
(33:18):
long journey so far?
Speaker 2 (33:20):
yeah, I mean there have been so many.
Jenny, it's like you, you haveto be open to that learning,
otherwise you will just notsurvive.
So I did mention this one ofgetting out of 3D printing
exclusively.
We still support customers in3D printing and we love those
applications, but we're morebroad than that.
(33:41):
I think you know, about a yearago we had a major moment where
we really understood ourtechnology as a shape analysis
tool.
So our technology is not aboutgenerating new shapes, and we
had been trying to generateshapes for four years and, by
the way, we created lots ofamazing shapes and helped people
(34:03):
design lots of great stuff.
But ultimately I think westarted to realize that the true
gap in the market wasn't inthat creation of new things.
It was in the understanding ofwhat those things do for a
business, you know.
Are they going to perform?
What is the price?
Are they similar to otherthings, like what is going to be
(34:25):
the manufacturing yield basedon tolerances?
These are business criticalquestions that come from
understanding shape, and sothat's a business insight.
At the same time, I think wekind of finally accepted that
our technology, its truestrength, was totally aligned
with that.
So its strength was not innecessarily design, although we
(34:48):
can actually do like tons ofcool design stuff but it's real
strength is, in this analysis,capability.
So that, I think, was kind ofthe key aha.
And then, if I can say one more,I mean I could like fill the
next half hour with this, butthe next one that I would say is
on the API point.
I want to ask you that actually,yes, yeah, yeah, so we did have
(35:11):
, and some people will know thatwe had this app in the market
and the app was for designingshapes and it was very cool.
Like I still use it actuallyit's quietly on the side but
what we came to realize is thatit's very hard to build an app
(35:32):
that is going to satisfy theneeds of all the people working
in 3D.
Uh, and even if we're justtaking the people working in 3D
printing, those people are inaerospace, they're in footwear,
they're in medical, they're inindustrial flow processing like
they're all over the place, andthere's no one size fits all
(35:53):
approach that we found anywaythat would do this.
And so what we realized is that, by providing the API, we were
able to provide much moretargeted solutions for our
customers, and so this alsocomes along with a change in
(36:13):
business model, that we engagevery deeply with our customers.
We don't have that manycustomers and we have bigger
engagements with a smallernumber of them and we make
technology that absolutely fitswhat they need and moves the
needle for their business.
So it's not about trying to dothat a little bit for everyone,
but rather get really focusedwith the API so that we can
(36:34):
deliver a very bespoke andtargeted solution.
Speaker 1 (36:37):
Do people still have access to the original version,
the cloud version, because Istill want to play around with
it, if that's possible.
Speaker 2 (36:44):
We have secret access , so, jenny, maybe we can
arrange something for you.
Speaker 1 (37:01):
Another question I have.
From the conversation aboutAPIs, you know, I just
discovered there is a smallindustry, probably also hidden
from the public, that focus ongeometric kernels.
The Russian governmentapparently owns a geometric
kernel that may or may not beaccessible to people.
What do you think of thatindustry and how are you guys
related to that?
Is it?
Speaker 2 (37:18):
the same thing or different.
Speaker 1 (37:23):
Yeah, such a fascinating topic and when you
look, this is on Wikipediathere's a list of geometry
kernels.
Speaker 2 (37:29):
This is how I learned about it.
Yeah, yeah, it is a short list.
A's a list of geometry kernels.
This is how I learned about it.
Yeah, yeah, it is a short list,a very short list.
Yeah, once you kind of strikeoff some of the more niche
kernels, it's a very short listindeed.
So when you think about that,it's remarkable that there's
maybe a handful like maybe fivekernels that drive all of the
CAD software, all the CAEsoftware that drive all of the
CAD software, all the CAEsoftware, like everything to do
(37:49):
with 3D geometry is basing itsform on one of those underlying
kernels.
So very, very interesting tothink about that as a business
and also to think about that ashow it has shaped the things
that we make in the world.
Like our tools shape us, so thetools we have access to shape
what we we make in the world.
(38:09):
Like our tools shape us, so thetools we have access to shape
what we've made in the world,and those kernels are
underpinning all of it.
So, to answer the second partof your question, yeah, metafold
, we have effectively built ourown geometry kernel, so it's an
implicit geometry kernel.
We are not dependent on any ofthese other geometry kernels,
but we do interface with thembecause that's an essential part
(38:31):
of um.
Speaker 1 (38:32):
You know, we need to be able to speak the same
language and not every one ofthem actually license their
kernels out either, and but onthe other hand, there are
probably some geometric kernelsout there that belongs to some
company that we don't reallyhave access to or know about.
Speaker 2 (38:50):
Yeah, this could be.
I don't know the story, by theway, with the Russian geometry
kernel.
Speaker 1 (38:54):
It's sort of like linked in folklore for me.
Yeah, it's fascinating.
I don't trust everything that'son Wikipedia, but I think the
general information on it isaccurate mostly, but some of the
stuff that's like totally outthere, um, you require some deep
digging, uh, and also trueunderstanding of the space,
(39:17):
which I do not have.
Um, all right.
So, um, in terms of 3d printingI know you're not entirely
focusing on it I would sayprobably it's accurate to say 3d
printing industry benefits morefrom metafold.
The metafold benefits from 3dprinting.
Is that right?
Speaker 2 (39:38):
maybe.
I mean, I think it's just thatas a business we couldn't
sustain ourselves in thatindustry, but we're still very
keenly interested and happy tosupport people.
Speaker 1 (39:45):
Yeah well, the reason I mentioned is that I think I
want people in the 3d, the 3Dprinting industry to understand
some of the unique benefit ofshape analysis, because they
probably have a certain type ofgeometries that would benefit
more than the mesh-basedtraditionally.
Totally yeah.
So do you want to unpack alittle bit of how your geometric
(40:07):
analysis would benefit 3Dprinting?
Speaker 2 (40:11):
Yeah for sure.
So 3D printing I mean, thereason we even started in 3D
printing is because thegeometric constraints are so
open here, right?
So people probably haven'theard you have free complexity.
Well, this is sort of true inthe sense that you can probably
print just about anything, butcan you actually represent it in
(40:33):
CAD?
And then, more relevant to whatMetafold does, can you
understand it before youmanufacture it?
And so I think, yeah.
So maybe this is the right timeto kind of get into lattices and
metamaterials.
Yeah, this is the and this willbe known to many of your
(40:54):
listeners but one of the coolthings about 3D printing is that
we can create lattices andmetamaterials.
So metamaterials are a way ofgetting new and different
material behavior just using onematerial.
New and different materialbehavior just using one material
.
So your three-dipinter isspitting something out and then,
by controlling the geometry andthe way that it packs together,
(41:15):
we can get really interestingand new material behaviors out
of that.
I honestly believe this is sucha huge area of potential, but
we're kind of somehow not readyfor it.
But I'm excited for the daywhen this can be more broadly
embraced.
But coming back to implicitgeometry and metafold one of the
(41:39):
great things about implicitgeometry and signed distance
fields and signed distancefunctions is that we can
represent very complexstructures very efficiently
Because remember, it's just afunction.
So when you have a function,it's very different from having
a surface representation, like atriangle mesh, where you have
(41:59):
to record the details of eachand every triangle that is,
mapping that very complexsurface.
So people are familiar with theheadaches of large file sizes
and processing time and all thatstuff, and this is where
implicit geometry really shines.
You can both represent thatgeometry but, relevant to
Metafold, we can also simulateit and we simulate it in the
(42:24):
same representation.
So there's no need to take yourimplicit representation and
tetrahedralize it and put itinto a solver.
We can do this all in one go.
So so Metafold and implicitgeometry more generally is a
great way to understand thekinds of shapes that you want to
produce with 3D printing.
Speaker 1 (42:46):
Yeah, um, you said that meta material is too early
of a project for you guys totake on.
Kind of now, I kind ofunderstand why you're pulling
out of the design creation partof the CAD process or the design
process, because maybe we justhave too many ideas.
Speaker 2 (43:07):
Yeah, I mean, there's still space for this, for sure,
and there's some good playersalready doing this kind of work,
I think and we continue, by theway, to support people who are
using metamaterials and 3Dprinting space for this, for
sure, and there's some goodplayers already doing this kind
of work um, I think and wecontinue, by the way, to support
people who are usingmetamaterials and 3d printing
like it's, it's still an activearea.
Speaker 1 (43:20):
It's just that it's not like a market yet and is
there any significant differencein computing power between the
mesh basedbased representationand the signed distance field
representation, these twodifferent ways?
Speaker 2 (43:38):
Well, yeah, I mean it's very compact.
The signed distance fieldrepresentation is very compact,
so you can end up with a veryconcise description of shape
that doesn't need like thesehuge file sizes.
You know there's still.
There's always a trade-offbetween accuracy and time.
You know, like with signdistance field, in theory
(44:02):
they're as precise as you wantthem to be, but in practice you
have to choose a representationto evaluate the function at.
So that choice ofrepresentation will depend on
compute, the function at.
So that choice ofrepresentation will depend on
compute.
It's very fast, but as you getmore and more detailed, as you
get more and you want to dothings bigger and bigger, you
will pay a certain time cost.
But when comparing to meshes,it's very performant, is there?
Speaker 1 (44:27):
such a day in the future is possible to figure out
what part of a component useSDF and what part use mesh as a
combined product?
Speaker 2 (44:39):
I think this is a great question and I think there
are some initiatives in thisdirection.
It's not what we've chosen tofocus on, but I think that is
the right question to ask,because there are certain
structures that are very wellrepresented by conventional,
conventional CAD representationsand there's some that are
better in implicit and so how doyou bring them together to find
(45:00):
this kind of like happy, happycombination really?
Speaker 1 (45:05):
and maybe also large amount of existing pre-existing
mesh that's out there that canuse your software to analysis.
Now coming to the more, perhapsa little bit futuristic and
exciting topic, which iseverybody is talking about AI
and machine learning these days,even though these topics really
existed for 40, 50 yearsalready and because there's
(45:29):
money pouring into these names.
Can you talk about how you seethis kind of wave of excitement
behind AI, ml and relevant toMetafold?
Speaker 2 (45:45):
Yeah, it's a good topic.
I mean maybe like the firstthing to say which I find people
are often confused about butMetafold is not an ai platform.
Right, we are just not.
It's not based on largequantity of data.
We're not training models.
We're based on that.
So that's the first thing tounderstand, but I think,
(46:06):
building on that, um, what'sinteresting is is coming back to
again, like our reason forbeing.
We're converting geometry intoinformation and ultimately, that
information should be the inputto any machine learning models.
So we certainly work withcustomers who are building
machine learning models I wastalking about the costing models
(46:27):
, for example.
This is an ML technique and weconvert three-dimensional data
into a feature vector, somethinglike literally a vector of
numbers that a machine learningalgorithm can ingest very easily
.
So the bigger point here, andsomething I've been thinking
about for a really long time, isthat AI and 3D do not go
(46:52):
together very well, and I evenclaim and people will, I think,
probably push back in some waysbut AI for 3D is not there, it's
not happening yet and it mightbe indeed a very long time for
it to happen.
There's a few reasons for this,but if you think about the LLMs
(47:15):
, for instance, the volume ofdata and the simplicity of that
data really is tremendous.
We don't have anything likethat for 3D.
And not only that the data we dohave tends to be proprietary.
There are very few, a verylimited amount of public 3D data
(47:38):
.
So this is one of only one ofthe challenges but, like the
other challenge is that humans,we understand the 3D world so
naturally and so easily.
And think about the last timeyou interacted with an LLM and
it made a comical error that AIdoes not understand anything.
(48:00):
So thinking about how you wouldeven extrapolate that level of
intelligence quote, unquoteintelligence to 3D when we don't
have enough data and it's justa more complex space, it seems
very, very difficult.
So I mean, there are subsets ofcertain applications and
(48:21):
subsets and very targetedproblems where ML can be very
impactful with starting with 3Ddata, but it's a very small
minority of cases.
Speaker 1 (48:32):
I would say it sounds like Metafo basically generated
raw data for some of the dataprojects of starting with 3D
data, but it's like a very smallminority of cases.
I would say it sounds likeMetafo basically generated raw
data for some of the dataprojects of various companies,
but you do not own those data,unfortunately, so you can't
actually create your internalmodel for any kind of future
applications, unfortunately.
Speaker 2 (48:50):
Yeah, I mean for what it's worth.
We are working on this.
We are actually collecting adata set.
One of our current projectsinternal R&D projects is
collecting a data set so that wecan build our own models, just
so that we have.
Yeah, as you say, everythingwe've done is always on
proprietary data, so if we ownour own data, then we can just
talk about the results a littlebit more clearly.
But it's very limited and it'shard to find the data.
(49:14):
It's really surprising maybe notbecause this is a really new
field is that we don't have anykind of open source for the
there's a little bit and there'ssome great data sets that have
been created by the academiccommunity and we love them and
we use them, but it's sort ofnot.
It's not enough if you thinkabout again, the quantity of
data that's feeding these verylarge models that are having
(49:35):
this big transformative impacton everything.
All this AI stuff.
We think about it all the time.
It's something we, as atechnology company, we have to
keep our eye on, but I kind ofalways come back to the same
point, which is that there'sstill a strong need for very
fundamental mathematicalrepresentation and results on 3D
(49:59):
geometry.
That will be the first steptoward any kind of meaningful AI
.
Speaker 1 (50:05):
Yeah, it's like building the foundation of
future AI.
Yes, exactly, and reallyinterestingly, I think any kind
of technologies that aim tointeract with external world,
three-dimensional world, willhave the same problem.
I read a really good articlewhich I can share in.
The link is about how robotics,the field of robotics,
(50:28):
especially the ones that we'reenvisioning, which is autonomous
, intelligent reasoning roboticsthose are not going to happen
either for any time.
So I think autonomous vehicleis probably going to be the
first one, and the major reasonfor that is creating a data set
that's useful for this, and thecost is already enormous
(50:49):
billions of dollars, decades ofefforts into it and it just
simply moved a needle so far.
And I would assume, like us, ourfield 3D printing or 3D
geometry, this kind ofinterfacing with external world.
We have the same data challengeand software challenge,
(51:11):
essentially to really envisionthat autonomous, automated
process.
So, yeah, well, fascinatingworld.
I hope the process isaccelerating instead of a linear
progression, yes, but rather anexponential field.
But the good news is to buildup that foundation, metafold has
(51:33):
tons of business yeah, well,yes, I mean, I sure, I sure hope
so so, um, okay, so we'rereaching the end of the podcast.
I have a couple of fun questionsfor you, okay, um, what are
some of the surprising moments,or many, oh oh, my god, there's
(51:53):
uh, there's been a lot ofsurprises.
Speaker 2 (51:55):
I mean.
Um, gosh, where to even beginwith this?
I mean, I think I thinkhonestly, I've kind of already
said the biggest surprise, whichwas understanding our
technology for analysis overdesign, because we weren't I
don't know somehow this one wasnot on our radar, like and it's
all in kind of the framing too,like our technology didn't
change from having that thought,you know, before and after, but
(52:19):
our understanding of itcompletely changed.
So this was a big surprise.
Um, there's surprising moments,I don't know.
There's more kind of comicalthings, like, in fact, when we,
when we began, we actually builta 3D printer and that's how we
got started as a company.
We were building this reallycool 3D printer and it was so
great and we were talking toinvestors and we had this one
(52:42):
set of investors who was veryinterested and we were building
the software and the hardware atthe same time and they were
very interested and they saidyou know, I think your strength
is in the software.
Like you've done something veryinteresting in software.
We're not as interested in thehardware because there's like a
lot of 3D printers on the marketand me and my two co-founders
(53:04):
went away and you know like ourhearts were in this printer.
You know, you get so attachedto these ideas and so we had
this big soul searching sessionand then we're finally like,
okay, we'll just do the software.
So we kind of got our acttogether on the software.
We went back to the sameinvestors and said like, hey, we
did the thing where you'repivoted, we're just doing
software now.
And they were like, oh, this isso great, this is so great.
(53:24):
We, um, we don't invest insoftware.
So I mean and there's been a lotof a lot of things like this,
you know, just um yeah, it'sit's just a journey and and I
think all the, all the little,the big and small learnings have
have kind of been surprisingalong the way you will think
(53:46):
that investor will tell youbefore.
I know, I know, yeah, it feltlike, okay, this is going to
unlock this next stage, butanyway, we were.
We were pretty nice.
Speaker 1 (53:55):
So many moons away.
I remember you joined one ofour virtual events and I think I
don't know if it's because ourpersonal conversation or maybe
the events you were talking tosomething about life, work,
balance and you're probably oneof the few founders actually
talk about that.
Not to mention, you're also oneof the few female founders who
(54:16):
also uses masks extensively.
So many unique features.
So how do you balance as afounder?
Speaker 2 (54:26):
I guess the first thing to say is I don't know if
I really believe in this idea ofbalance.
Speaker 1 (54:32):
Like if I'm honest about it so it's.
Speaker 2 (54:36):
it's not a balance per se, um, I think it's more
that I've I've learned enoughabout myself to set boundaries
and to know what I need to feellike a human being and uh, so
you know concretely this work wedo.
It's so you know, it's in ourminds, right, it's in our minds.
(55:00):
It's not embodied in any way.
It's, yeah, it's easy to kindof get stuck in that world, and
so I really think a lot about,just like, movement I'm an
absolutely obsessive exerciser,this keeps me sane but also
things like handwriting, youknow, getting off the computer
and and handwriting, um, so thekinesthetic action of, of
(55:24):
writing on a piece of paper witha pen, um, and similarly,
drawing.
So I'm a big believer thatwriting is thinking and drawing
is seeing, and so when I want tothink, I write and when I want
to see which is well, I alwayswant to see, but I just I like
to draw and I think that's animportant practice as well for
(55:45):
me.
Kind of quote, unquote balanceis like, well, drawing
boundaries around my time andthen making sure I have time to
like be a human, a physicalhuman, in the world.
Speaker 1 (55:59):
Yeah, I have to say you're really a unique person as
a combination of artists butalso mathematician, which
basically is a concrete set oflogics, are actually quite
inflexible.
In a way.
You have to say one plus one isequal to two is absolute truth
and several steps of logic goingtowards a certain outcome.
(56:21):
So it's, it's fascinating tosee both features in one.
I have to say my, uh, my brainis malfunctioning, short
circuiting at the moment.
Speaker 2 (56:31):
Well, no, but I, I but I, I mean I might push back
on that a little bit, becauseone of the I'm actually terrible
at arithmetic, like terrible.
Um, that's not the way my brainworks, but I think, uh, for
mathematics, I think somethingpeople miss about mathematics is
that it's very creative, verycreative.
So this is really like thelinking theme between between
(56:56):
what I do in mathematically orartistically or whatever like,
and and also what I do as abusiness person, like I'm.
I'm interested in facilitatingcreativity in our customers to
allow them to achieve thesethese, uh, these creative
engineering outcomes basically.
But I think that is for me whatI've come to as the linking
theme between mathematics andart.
Speaker 1 (57:17):
Yeah, I mean.
Some people said thatmathematics is nature's language
.
Speaker 2 (57:21):
It's universal.
Speaker 1 (57:23):
So well, just for fun , what do you read or observe or
listen to to get inspirations?
Any kind of external mediasource that our listener can
also try to go after?
Speaker 2 (57:43):
Yeah, I mean I try to read broadly, like it's easy to
get into, like the way ourmedia landscape is, it's easy to
kind of get into this um kindof place where you're just
receiving content that's kind oflike tuned to you.
So try and get out of that alittle bit and I read like more.
Let's say, in the past fiveyears, become more and more
(58:05):
interested in, in readinglong-form books, like just
actually books, not long booksbut just books, and the reason
for this is that, like so muchof the information we receive is
very short, like very shortform, and actual ideas, like
real ideas, are messier andcomplex and need maybe a
book-like thing to really getinto.
(58:26):
So I try to read real books.
I read them very slowly, soI'll just put it like this I
haven't read that many books,but I can maybe give one
recommendation of a book I readrecently which again is like
unrelated to anything else, butit's called Underland.
It's a nonfiction by this author, robert McFarlane, and he's a
(58:52):
nature writer but also kind of aphilosopher and a poet.
So it's beautiful.
It reminds you of why theEnglish language is, uh, is
interesting, it's very rich.
He like expands my vocabularywhen I read him.
But Underland is all about ourrelationship with everything
under the surface of the earth.
(59:13):
So it's about I think thesubtitle of the book is called A
Journey Into Deep Time and itputs kind of so much perspective
on our everyday busyness whenyou think about, like, how long
rocks have been there, forinstance.
But it goes much, much deeperthan this.
He looks at all kinds ofunderground exploration and the
(59:35):
idea of burial and just ourrelationship with the surface of
the earth.
It is fascinating and range andbeautiful and uh, and that's
the kind of thing I like to readis something that's like very
different from from my, my, myday to day and uh, and provides
like a different context.
Speaker 1 (59:56):
And a bigger view of the world and life Way bigger.
Yes, exactly, okay, and Ipromise you.
This is a one last question Doyou have any advice for our
young listeners?
Speaker 2 (01:00:10):
Young listeners.
Well, too many.
Speaker 1 (01:00:14):
I don't know from from, uh, I think well, you know
, right now, I have to say, weare in this wave, or somebody
called bubble of ai, and a lotof new graduates actually in
computer science and mathematicscan find jobs.
Speaker 2 (01:00:34):
So this is the rumor, I don't know.
Yeah, so maybe you have ourdata, jenny.
I don't know.
Speaker 1 (01:00:45):
I read something in Wall Street Journal just
recently saying that the lowertier computer science engineers-
hiring is decreasing.
Speaker 2 (01:00:49):
So yeah, yeah, so.
Speaker 1 (01:00:49):
I believe this.
Speaker 2 (01:00:50):
However, I would draw a pretty firm distinction
between that and getting amathematics education.
So I think I guess my advicewould be study mathematics.
Yeah, not necessarily becauseyou're going to become a
(01:01:11):
mathematician, but rather thatit's a.
It's a discipline for how tothink and how to reason through
arguments, and it teaches you acertain way of of approaching
the world that I think isfocused on searching out truth,
which is something I really careabout, and and you know asking
(01:01:31):
questions like asking reallyfundamental questions, that, and
you know asking questions likeasking really fundamental
questions that can yieldinsights.
And so, yeah, study mathematics, become a problem solver, and
these skills are indispensable,no matter where you go.
Speaker 1 (01:01:46):
Absolutely.
I love that advice and,honestly, the only class I
remember thus far is my mathteacher, is my mathematics
classes In a good way, johnny,in a good way, and not just
learning how to do things, buthow to solve problems in many
different ways creatively.
Thank you so much for joiningus today, Lisa.
(01:02:08):
I hope to see you soon again.
Speaker 2 (01:02:10):
Yeah, it's been my pleasure.
Thanks for the invite, jenny.
It's been a great conversation.
See you soon again.
Yeah, it's been my pleasure.
Thanks for the invite.
Speaker 1 (01:02:15):
Jenny, it's been a great conversation.
See you next time.
This podcast is for educationaland informational purposes only
.
The views expressed do notconstitute medical or financial
advice.
The technologies and proceduresdiscussed may not be
commercially available orsuitable for every case.
Always consult with a licensedprofessional.
Hello there, welcome to the Lattice podcast, episode
number 93.
Today, we explore the worldwhere mathematics meets
manufacturing innovation withElisa Ross, co-founder and CEO
of Toronto-based startup Metafo3D.
Elisa and her team aretransforming the way industries
analyze and understand 3Dgeometry, using advanced
(00:24):
geometric intelligence andimplicit modeling.
Whether you are a designer,engineer or anyone fascinated by
the creative side of math, getready for an eye-opening
conversation about shape and thefuture of making things.
Please listen to the disclaimerat the end of this podcast.
Please listen to the disclaimerat the end of this podcast.
(00:47):
Hello, hello.
Thanks for joining the pod.
Today we have the CEO andco-founder of Metaphone 3D,
elisa Ross.
Hello Well, elisa.
I feel like today's interviewis a bit like a challenge for me
because, even though I lovemath, this is like one of those
mathematical challenge Iencountered when I was in school
(01:08):
.
It's both exciting and kind offascinating, but also I
understand the limitation of myknowledge.
So first of all, I'd love tojust hear a little bit about
what Metafold 3D is and a littlebit about yourself what
Metafold 3D is and a little bitabout yourself.
Speaker 2 (01:30):
Yeah, absolutely Happy to Well.
First of all, thank you, jenny,for the invitation.
I'm super excited to be hereand I promise we won't go too
deep on any mathematical topics.
Metafold 3D is a startup.
We're about five years old now,we're based in Toronto and
we're all about math.
As you've already introduced,I'm a mathematician by training,
specifically geometry, and wecan get into that if you like.
(01:52):
But Metafold is all aboutbringing geometric insight to
manufacturing, and our productis called Geometric Intelligence
, and it's really all aboutturning geometry into
information that can be used tomake decisions.
We can unpack all of that, butthat's the highest level.
Speaker 1 (02:16):
Awesome, I love it.
Yeah, I think we want to unpacksome of the maybe the most
important concepts of yoursoftware, because I am also
learning as we are doing thisinterview, and why don't you
unpack some of the mostimportant concepts at the core
of your software?
Speaker 2 (02:37):
Sure.
So geometric intelligence isour technology and it's really a
platform for shape intelligence, to shape information and
analysis.
So what I mean by that is thatwe take geometry as input and we
give people using our softwareinformation about that geometry.
(02:57):
So why would you want this?
The idea is really that youknow, shapes are not that
meaningful on their own when wemanufacture things.
These shapes exist in the worldto perform a purpose, to serve
a purpose, and so our approachis all about helping people
(03:20):
understand how their shapeswould perform or how they will
be manufactured in the worldbefore they go ahead and make
those things.
So, very specifically, weconvert geometry into
information about performance.
Like you know, how well will itperform under certain loading
(03:41):
conditions?
Let's say, we convert it intomanufacturing information.
We also do a similarity.
We can look at similarities, soanswer questions like have I
made something like this in thepast?
And we can also address thingslike tolerances, which have a
huge impact on manufacturing atscale.
So, again, we can get into allof those topics, but the thing I
(04:04):
really want to focus on oremphasize is that we convert
geometry into information thatcan be used to make decisions
Under the hood, of course, andwe can get into the mathematics
of it under the hood.
We take a fairly differentapproach from traditional CAD
software.
Speaker 1 (04:22):
Yeah, I want to unpack the concept of CAD before
because I had the samemisconception about it, since I
am not a designer and I don'twork with it day in, day out.
But CAD software itself doesinclude analytics and also
provides some kind of analysisfor manufacturing and outcome.
(04:44):
I mean, some of the biggercompanies that I have in mind
are Autodesk or SynopsysSynopsys sorry they are.
They also have designcapability and analytics
capability.
Where do you guys are you?
Are you guys part of this chainof information or you are
(05:07):
completely different from that?
Speaker 2 (05:10):
Yeah, you're right, all of these platforms do
provide some kind of analysiscapability, and the reason is
what I stated before, that theshape is kind of useless on its
own.
You need to know something aboutthe shape.
Where we really differ fromthose platforms is in the way we
represent geometry, and thatgoes into the details of our
technology.
But, in short, we representgeometry using an implicit
(05:35):
approach which is different fromthe way traditional CAD is
represented, and this allows usto extract different kinds of
information and do thingsdifferently.
So it means that we have, forinstance, a faster path to
simulation, and so if you'redoing simulations one at a time,
(05:58):
then conventional CAD isprobably a great choice.
If you're trying to do athousand simulations at once,
then maybe we should considerhaving a conversation.
This is the kind of thing wecan help with.
So that's actually.
Another component of ourtechnology is that we are really
built around working with largevolumes of 3D data.
(06:18):
So if you have a lot ofgeometry, you're trying to do a
lot of experimentation or you'regoing to produce a lot of parts
, then this is a very good spotfor our technology to slot in,
because if you use conventionalCAD, you're probably looking at
those things one at a time,which is very time-consuming.
Speaker 1 (06:38):
And yeah, I have to say this is a very enlightening
conversation for me personally,because decades ago, believe it
or not, I actually did researchin simulation of a 3D shape
which is aneurysm, because wewant to use a mesh to predict
what kind of aneurysm shapewould rupture.
And actually that kind ofinformation now translated to a
(06:59):
company called the HeartFlow anda bunch of other
medical-related simulation toolsto predict heart attacks and
aneurysm, ru related simulationtools to predict, you know,
heart attacks and aneurysmrupturing medical device, but
they're all based on thetraditional CAD.
Now I thinking, thinking back,and I remember I have to wait
for a whole day for the analysisto come in, for the stress
(07:19):
sheer stress calculation and,yes, just to create the mesh
itself takes a couple hours andwe're talking about circa 2010.
And I remember that I work withPhD, so like I will like figure
out the shape of the aneurysmto be simulated, and then he
would keep the computer onovernight to generate something
(07:41):
the next day.
It's either a fail or success.
It's tremendous amount of work,but the results you know.
It just takes so long.
I remember that had we hadimplicit modeling capability,
things would be different andmaybe we can test 10,000
different aneurysm models.
Speaker 2 (08:00):
This is possible.
Yeah, I mean it's also possiblethat that approach was actually
great and that it was just fineIn some cases, depending on the
type of simulation you'rerunning, and then they need to
change the format of the datainto a different format for
(08:27):
simulation.
Maybe it's a different person,that's what you described.
There's someone else whoactually runs the simulation and
so, you know, in largecompanies this might be a fully
separate team.
So if you're trying to iterateon a design and you're doing
that super slow process, thisjust takes too much time.
What we can offer is helpingpeople bring a lot more of that
(08:48):
information to the designerreally actually in some cases so
that they can run theirsimulations and get information
that can inform their designprocess much more immediately.
Speaker 1 (09:00):
So are there certain type of shapes in particularly
would benefit from Metafo'ssoftware.
Speaker 2 (09:09):
I wouldn't say there are certain types of shapes.
No, I think what's kind ofinteresting to understand is
that so much of the way weinteract with geometry on a
computer will come down to howyou represent geometry.
And so for us it's not so muchthat there's certain shapes that
are more amenable to approachthan others, it's more kind of
(09:33):
levels of accuracy, let's put itlike this, and depending on
what you're trying to achieve,you might choose different
representations.
So for manufacturing, youprobably need a traditional
representation like a B-rep oreven like a test-related format,
like the STL probably familiarto you in the 3D printing world.
(09:54):
But for what?
Our perspective at Modifold isthat if you're trying to get
information about the shape, itcan be very helpful to move to
an implicit representationbecause you have all this
information about the shape.
It can be very helpful to moveto an implicit representation
because you have all thisinformation about the volume of
the structure and you can dooperations on it that give you
information that you can't getwith a traditional format.
(10:15):
So it's not so much about thetype of shape as it is maybe in
the kind of information you'retrying to extract from your
geometry.
Speaker 1 (10:23):
Alyssa, I view you as my mathematic teacher, so there
are two concepts I'm hopingthat we can unpack for average
people who listen to our podcast.
One is implicit modeling, whichis a new concept to me, and the
other one is sign distancefunction.
I know they sound like bigwords, but actually in reality,
after a while, I feel likethey're actually not that
(10:45):
complicated to understand.
Speaker 2 (10:47):
Yes, I agree, and actually they're they're really
one in the same in a lot of ways.
So let's start with the signeddistance function.
This is a very um, simplemathematical function and it it
operates on three-dimensionalspace.
So you begin with a shape, asurface in three-dimensional
(11:07):
shape space rather, and the sinedistance function is defined by
that shape and for every pointin three-dimensional space it
returns the distance from thatpoint to the surface of the
shape.
So this is a function that iszero, I would say.
In mathematical speaking.
(11:28):
We would say identically zeroat the surface of the shape.
Everywhere else it is non-zeroand is actually equal to the
distance from the shape.
If you're outside the shape,it's positive, if you're inside,
the convention is that it isnegative.
So already this gives us such arich space of information that's
(11:49):
not present in, let's say, atriangle mesh.
So for comparison, a trianglemesh is a set of triangles that
you stitch together toapproximate the surface of a
geometry.
Important to understand in bothcases.
This is not geometry, it's arepresentation of geometry, and
(12:11):
I like to make that pointbecause people often have kind
of a mental idea of whatgeometry is, what it looks like,
and that is always approximatedon a computer.
So the sine distance functionis a precise and complete
definition of the shape.
But when we represent it on acomputer we probably need to
(12:32):
sample a grid to, like you know,sample the distance across the
grid of values to reallyunderstand where that shape is.
So that's the concept of signdistance function, Clear so far.
Speaker 1 (12:46):
Yes, so the mesh and the sign distance function.
Would it be accurate to saythey're both kind of ways,
different ways to represent a 3Dgeometry?
Speaker 2 (12:59):
Absolutely, absolutely.
And there's a third way.
There's multiple ways, but athird common way is the boundary
representation format, the BREP, and this builds up surfaces
through points, lines andsubsurfaces basically too
(13:20):
listeners.
Speaker 1 (13:22):
Well, so I want to give us some real-life examples,
if you're allowed to, or maybesome hypothetical examples, of
how Metafold works, as opposedto, let's say, autocad and all
the other CAD software that hassome kind of intrinsic analytic
(13:43):
capability, and also for peoplewho are already working with
those kind of softwares can theyuse your program seamlessly,
kind of integrated?
Speaker 2 (13:49):
into it.
Yeah, these are.
These are good questions, um,maybe just because I want to
close the loop on this.
You did ask about implicitmodeling.
Yeah, I want to say thatimplicit modeling is just refers
to using signed distance fields, usually to build up more
complex geometric shapes.
So, like conventional modeling,you're usually, you know, in a
(14:12):
boundary representation orcreating a triangle mesh, and
when you're doing implicitmodeling, you're using some kind
of volumetric representation,probably a signed distance field
, and this is actually kind of abeautiful thing.
If you think about signeddistance fields, you can
represent them as a grid ofvalues.
So if you, if you want to say,add two together, it's as simple
(14:33):
as arithmetic on that grid.
So certain operations becomemuch simpler in implicit
modeling world, which is whypeople tend to like it.
So, from uh, from that to yourquestion about real world
applications.
Yeah, so I'd love to give someapplications and I, given your
(14:54):
audiences is in the health space, I'll try and keep them kind of
health care adjacent.
Um, and actually, on that, oneof the the big application areas
for us is in performancesportswear, specifically
footwear, so running shoes, andhere the goal is to create the
(15:15):
most performant running shoespossible.
So we work with teams who areworking on that problem, that
problem.
There's been many materialdevelopments over the past
couple of years and also justdevelopments in the way the
materials and geometry are puttogether that have created
pretty remarkable outcomes interms of athletes, like we've
(15:37):
seen the marathon records beingbroken and all sorts of other
similar records being broken,largely by footwear technology.
So there's a real interest inpushing forward this technology.
The standard process forproducing footwear is to come up
with an idea, physicallyfabricate it and test it.
(15:58):
Okay, so this is certainly agood way of getting information
about geometry and material.
However, it's very slow, soteams will spend months in these
prototyping cycles trying toiterate and find the best
performing shoe, but in fact,it's not an exhaustive search
(16:19):
because that is just such atime-consuming process.
So what we help with here istaking a broad sweep of geometry
, taking a material palettethat's potentially specific to
that manufacturer, and thenhelping them scale up their
(16:40):
experiments.
So really simply, all we do isreplicate their physical tests
digitally and then radicallyscale them up so that they can
simply get so much moreinformation than they would
through a physical or manualprocess.
That makes total sense.
Yeah, and that is.
(17:02):
You could imagine this sameprocess working in a more
healthcare related field, whereyou're trying to really, you
know, provide some level ofcushioning and you want to make
sure that you're achievingexactly the right response to
force, depending on theapplication.
But often we see the samecombination of geometry and
(17:26):
material.
You want to vary both of themsystematically so that you can
understand how your partperforms.
Speaker 1 (17:34):
Yeah, I mean, I can see this can definitely
translate, not just in the shoeindustry, but all kinds of other
wearables or pathetics.
What about implants?
Would that be?
If people want to iteratedifferent designs for hip
implants, for example, can theyuse your software to somehow
scale up their design process?
Speaker 2 (17:57):
Yeah, absolutely.
I mean, it's the same kind ofthing.
What is sort of interestingabout the implant world, though,
as I understand it, being anoutsider to the space is that it
is patient-specific, and sothat element might indicate that
you know you're probably for agiven patient.
You might not want to spend toomuch time, you know, physically
(18:20):
testing, and that mightactually make it a very good use
case for a small digitalexploration, like given the
patient geometry.
We're trying to find the bestperforming implant.
Here's a systematic way ofstudying this.
It takes, you know, an hour orwhatever.
It takes digitally, and then atthe end of it you find the
candidate that works the bestand will yield the best outcomes
(18:43):
.
Speaker 1 (18:44):
Well, as much as 3D printing industry would like to
have personalized implants, thereality is the majority of the
implants are not personalized,but more and more implants are
now 3D printed.
I think even just like onelevel up improvement in design
not personalized, but let's saymodel certain X or YZ can be a
(19:07):
better fitting for osteoporoticpatient and stuff like that Like
just incremental improvement.
They need design iterations aswell.
Speaker 2 (19:18):
Yeah, exactly, and that's a great point, that a lot
of these are based on kind ofbest practices as opposed to
lengthy iteration cycles to findthe optimal solution, or not
necessarily the optimal, but anoptimal solution.
So yeah, absolutely, that wouldbe a great fit.
Speaker 1 (19:36):
And if I'm not?
Speaker 2 (19:36):
this speaks to kind of like a bigger theme for us as
a startup, which is that wehave really moved away from
focusing on 3D printing.
So we, you know, when we began,we away from focusing on 3D
printing.
So when we began, we were veryfocused on 3D printing because
there are all kinds ofinteresting geometric things.
Of course that's how weconnected, but as a market it's
(19:59):
just not quite growing the waythat we believe it should and
hope it will.
But more than that, inconventional manufacturing,
whether you're talking aboutmedical or footwear or aerospace
, there are so many interestinggeometry problems and
interesting places to optimize.
So, yeah, but understandingperformance is one of our key
(20:20):
pillars of turning geometry intoinformation.
It's not the only one, so wealso work in instance or
on-demand manufacturing.
We work in kind of similarityanalysis for part reuse and also
tolerancing.
I don't know if any of thoseresonate for you, Denny, or want
(20:40):
to pursue further.
Speaker 1 (20:41):
I want to go back to the fact that you pivoted away
not away, but like expanded yourhorizon significantly because
you discovered market and wefrequently interview people who
did that, because I think that'sa natural progression of
knowledge and insights as anentrepreneur.
And it's actually good for you,because 3D printing industry is
(21:04):
a small industry and it stillis growing very fast.
But to survive as a company,you need to discover new markets
, you need to create the market,and I'm just surprised in a
good way that you have found amuch bigger market that is not
well served by the currentsoftware providers.
Speaker 2 (21:25):
Yeah, this is.
It's exciting for sure, and youhighlight a bit of a point here
, which is that the currentsoftware providers and CAD in
general is a very old and stickykind of set of technologies of
geometry we were talking aboutearlier.
(21:46):
They exist both because theyare pretty good, but also just
because they exist and they'vegot inertia and they've got
people who've got decades offiles in those formats.
So it is interesting to beapproaching this large market of
manufacturing with a newapproach to shape analysis and
(22:06):
geometry understanding, and wedo certainly believe that
there's a lot to be gained fromthis approach that's distinct
from what the kind ofconventional players are doing.
Speaker 1 (22:18):
Yeah, and also I feel like the market probably is
also slowly responding to youand your company's product
because, like you said, theyweren't aware of new ways of
doing things because of theinertia.
And how do you tell people youknow this way is a better way or
more productive for you?
Speaker 2 (22:38):
I mean, I think it's.
I think that's been kind of akey understanding for us too, is
that it's not really themessage is not hey guys, this is
the better way, it's more.
What more can we do withgeometry?
Ultimately, everyone wants thesame thing.
They are using geometry as atool to produce something, and
(23:03):
the last thing they want to bedoing is thinking about their
geometry representation.
They just want to design athing, they want to understand
how it's going to work andproduce it, and then obviously,
they want to make that veryprofitable.
And so to make it profitable,you ultimately need more
information, and so thisunderstanding from us.
(23:24):
It's not about changing the waypeople design things.
It's about taking the thingsthey're already doing and giving
them more information aboutthem so that they can make
better things, make fasterthings, increase their yield.
You know, stop redesigning thesame part over and over all
these kinds of things.
Speaker 1 (23:44):
I think that you're one of the companies that is
cloud native, the nextgeneration of software provider,
and now you're also going APIfirst strategy in terms of
business development.
And the fact is, it's not justgeometry analysis.
You also provide a suite ofbenefits like instant quoting,
(24:08):
tolerance analysis, like yousaid, and smart part reuse.
I'd love to hear you justunpack some of those features,
because I don't see this veryoften.
Speaker 2 (24:24):
Yeah, for sure, and I think like a couple of things
here.
So, cloud native yes, we have acloud offering.
However, we are not exclusivelycloud and this has been like
another sort of semi-painfullesson to learn right that if
there's certain industries, forexample, aerospace and even, to
some extent, sportswear, wherethey just like they don't want
to be on the cloud, and I get it.
So we have solutions for peoplewho want to achieve dramatic
scale using the cloud and wehave solutions for people who
(24:45):
want to run it on the computerunder their desk.
This is all possible.
But yeah, so in terms of theother offerings we have, it is
all rooted in the sametechnology.
So using signed distancefunctions to gain understanding
about shapes For the InstantCADquotation super interesting
(25:07):
problem.
Quotation Super interestingproblem.
And I think increasingly we'reseeing more and more
manufacturing that people aretrying to manufacture more
domestically for all thegeopolitical reasons.
But really interesting problemscome up in the on-demand
manufacturing space.
So the idea in that business isthat those on-demand
(25:29):
manufacturers they receive apart and they need to provide a
quote as fast as possible.
So they need to provide a quoteto the customer on how much
it'll cost the customer to buy,but they also need to have a
supplier cost.
They need to provide, you know,like put it out for their
suppliers to produce for acertain price.
So how do they get thatinformation?
(25:50):
Certainly, it's a function ofthings like material and lead
time and quantity and a wholebunch of non-geometric factors,
but at some point there's a lotof shape analysis that needs to
come in there.
So, understanding the part,what kind of machine do you need
to make it on and what are thefactors that would make it more
(26:12):
or less complex to produce?
How many hours would it take toproduce this?
So we don't do those things.
What we do is we convert thegeometry into a set of features,
geometric features and otherforms of data that those
companies can then ingest andbuild their costing model.
(26:34):
So I hope that makes sense.
We're not providing the price,we're not providing the process.
We're converting the geometryinto other forms of information,
other measurements, differentkinds of analysis results that
allow them to create models thatare accurate for pricing.
Speaker 1 (26:53):
Does it include manufacturing process materials
and that kind of information inyour geometry outputs?
Speaker 2 (27:01):
So we are really strictly on the geometry side,
so we provide the geometricfeatures.
They bring all thenon-geometric features and
that's why they are building thepricing model themselves.
But we can, you know, via ourAPI, just give them all of this
data about the geometry, andwhat this means is that fewer
(27:22):
parts where an expert needs toopen up the file, just one at a
time, spin it around, estimatethe time it would take to
produce it, which is kind of thestate of things at this point.
There's some automation andthere's some expert quoting that
has to go on.
So how do we increase theamount of automation for that
process?
Speaker 1 (27:44):
I definitely understand that you guys are
purely in the geometry business.
However, doesn't it soundlucrative if you are able to
create an end-to-end pricingquote instantaneously?
Speaker 2 (27:55):
Let me tell you this is quite a hard problem.
Yes, it is definitely lucrative.
And there are businesses thatare operating on this basis, for
sure.
Speaker 1 (28:06):
I think I can imagine why some of the manufacturers
would not want to share the datathat they use because it's
proprietary for their ownprocess.
Definitely, yeah.
And also, rome is not buildingone day.
Metaphode is not coming out ofthin air, so I want to just go
back to your original founderstory, if that's possible, to
(28:28):
tell us a little bit about yourearly days before Metaphone and
how that translates into today.
Speaker 2 (28:35):
Sure, I don't know how far back you want to go.
Speaker 1 (28:39):
By the way, just curious, I was reading the
history of cat, which I willshare a link in a podcast.
Are you related in any way withDoug Ross?
Speaker 2 (28:48):
I am not.
I'm not in any way related tothat person.
Okay.
Speaker 1 (28:52):
We invented the term cat, basically.
Oh, I didn't know that.
Speaker 2 (28:56):
Okay, well, maybe I should claim I'm related and
have more credibility.
Speaker 1 (29:00):
He invented the term computer-aided design in the
1960s and his name is Doug Ross.
Mit guy permanently imprintedhis name in the industry.
Speaker 2 (29:13):
Wow.
Well, I can aspire to that kindof goal.
I mean, there are a lot ofRosses in the world.
At this point I'd definitelyrather that metaphor make it
smart rather than me personally.
But so, yeah, I am not relatedto Doug Ross my background so
I'm a mathematician by training.
But it may be worth saying thatI never set out to be a
(29:36):
mathematician.
This is not what I thought Iwould be doing.
If you had told me this is theway things would work out when I
was a kid, I would be laughing.
I really wanted to be an artistand math was just not on my
radar.
I'm a very, very visual thinkerand so I got into math through
art.
I was really fascinated bytilings and patterns, these
(29:58):
two-dimensional patterns andkind of geometry stuff.
And then that interest carriedme through first an
undergraduate and the master'sand then finally the PhD.
But I've always been veryfascinated by the interaction
between geometry andapplications.
And it's a little unusualbecause I'm more interested in
(30:20):
pure math and applying ideasfrom pure math to problems in
the real world.
This is in contrast to thefield of applied math, which
always begins with a questioncoming from another field of
science usually, and there's awhole set of mathematical
techniques that fall under theapplied math, but that's not
(30:40):
really been my interest.
I've been interested in theoryof geometry and combinatorics
and things like this thatnevertheless have applications
to especially material science.
So my PhD was on, really, thestructure of zeolite materials
that have a very interestingmolecular structure and this has
led to an interest in latticestructures and metamaterials,
(31:03):
which has, you know, inspired alot of my work in 3D printing.
But so all that to saymathematician but kind of like
an uneasy relationship withmathematics in certain ways, and
very interested in art, indesign, and so, following, you
know, the whole academic path, Imet the person who's now my
(31:26):
co-founder.
This is Daniel Hamilton, whowas working.
He had started his ownconsulting business applying
mathematics to architecturebasically, and so, given my
interest, you know, we clickedright away and I started working
for him and then we workedtogether for quite some time and
(31:46):
eventually kind of became likea partnership in that business.
And then eventually, you know,we had a small team and one of
those team members, tomRoslinski, joined the three, the
two of us and the three peoplestarted Metafold.
So yeah, and Metafold I guess Ican say was inspired by the
(32:10):
consulting work we were doing.
So in this consultancy, likewho is it who is hiring
professional mathematicians tosolve problems?
So we mentioned architecture.
But we ended up working, infact, in sportswear for a
company that was trying to dosomething really ambitious using
3D printing.
And it was that context likethey're literally hiring
(32:34):
mathematicians because they haveproblems they can't solve about
3D geometry.
What are the like, what's goingwrong here and what's not
functioning about their existingsoftware for CAD, for designing
and understanding those shapes.
So that was the kind of themotivation for Metafold.
Speaker 1 (32:56):
And now it's been five years and I think I've seen
you guys grew at least threeyears now.
I've known you three years agomaybe.
What are some of the ahamoments for you?
Because I know you're very opento new ideas and ready to pivot
whenever, um, it's necessary.
Like, what are some of the ahamoments for you throughout this
(33:18):
long journey so far?
Speaker 2 (33:20):
yeah, I mean there have been so many.
Jenny, it's like you, you haveto be open to that learning,
otherwise you will just notsurvive.
So I did mention this one ofgetting out of 3D printing
exclusively.
We still support customers in3D printing and we love those
applications, but we're morebroad than that.
(33:41):
I think you know, about a yearago we had a major moment where
we really understood ourtechnology as a shape analysis
tool.
So our technology is not aboutgenerating new shapes, and we
had been trying to generateshapes for four years and, by
the way, we created lots ofamazing shapes and helped people
(34:03):
design lots of great stuff.
But ultimately I think westarted to realize that the true
gap in the market wasn't inthat creation of new things.
It was in the understanding ofwhat those things do for a
business, you know.
Are they going to perform?
What is the price?
Are they similar to otherthings, like what is going to be
(34:25):
the manufacturing yield basedon tolerances?
These are business criticalquestions that come from
understanding shape, and sothat's a business insight.
At the same time, I think wekind of finally accepted that
our technology, its truestrength, was totally aligned
with that.
So its strength was not innecessarily design, although we
(34:48):
can actually do like tons ofcool design stuff but it's real
strength is, in this analysis,capability.
So that, I think, was kind ofthe key aha.
And then, if I can say one more,I mean I could like fill the
next half hour with this, butthe next one that I would say is
on the API point.
I want to ask you that actually,yes, yeah, yeah, so we did have
(35:11):
, and some people will know thatwe had this app in the market
and the app was for designingshapes and it was very cool.
Like I still use it actuallyit's quietly on the side but
what we came to realize is thatit's very hard to build an app
(35:32):
that is going to satisfy theneeds of all the people working
in 3D.
Uh, and even if we're justtaking the people working in 3D
printing, those people are inaerospace, they're in footwear,
they're in medical, they're inindustrial flow processing like
they're all over the place, andthere's no one size fits all
(35:53):
approach that we found anywaythat would do this.
And so what we realized is that, by providing the API, we were
able to provide much moretargeted solutions for our
customers, and so this alsocomes along with a change in
(36:13):
business model, that we engagevery deeply with our customers.
We don't have that manycustomers and we have bigger
engagements with a smallernumber of them and we make
technology that absolutely fitswhat they need and moves the
needle for their business.
So it's not about trying to dothat a little bit for everyone,
but rather get really focusedwith the API so that we can
(36:34):
deliver a very bespoke andtargeted solution.
Speaker 1 (36:37):
Do people still have access to the original version,
the cloud version, because Istill want to play around with
it, if that's possible.
Speaker 2 (36:44):
We have secret access , so, jenny, maybe we can
arrange something for you.
Speaker 1 (37:01):
Another question I have.
From the conversation aboutAPIs, you know, I just
discovered there is a smallindustry, probably also hidden
from the public, that focus ongeometric kernels.
The Russian governmentapparently owns a geometric
kernel that may or may not beaccessible to people.
What do you think of thatindustry and how are you guys
related to that?
Is it?
Speaker 2 (37:18):
the same thing or different.
Speaker 1 (37:23):
Yeah, such a fascinating topic and when you
look, this is on Wikipediathere's a list of geometry
kernels.
Speaker 2 (37:29):
This is how I learned about it.
Yeah, yeah, it is a short list.
A's a list of geometry kernels.
This is how I learned about it.
Yeah, yeah, it is a short list,a very short list.
Yeah, once you kind of strikeoff some of the more niche
kernels, it's a very short listindeed.
So when you think about that,it's remarkable that there's
maybe a handful like maybe fivekernels that drive all of the
CAD software, all the CAEsoftware that drive all of the
CAD software, all the CAEsoftware, like everything to do
(37:49):
with 3D geometry is basing itsform on one of those underlying
kernels.
So very, very interesting tothink about that as a business
and also to think about that ashow it has shaped the things
that we make in the world.
Like our tools shape us, so thetools we have access to shape
what we we make in the world.
(38:09):
Like our tools shape us, so thetools we have access to shape
what we've made in the world,and those kernels are
underpinning all of it.
So, to answer the second partof your question, yeah, metafold
, we have effectively built ourown geometry kernel, so it's an
implicit geometry kernel.
We are not dependent on any ofthese other geometry kernels,
but we do interface with thembecause that's an essential part
(38:31):
of um.
Speaker 1 (38:32):
You know, we need to be able to speak the same
language and not every one ofthem actually license their
kernels out either, and but onthe other hand, there are
probably some geometric kernelsout there that belongs to some
company that we don't reallyhave access to or know about.
Speaker 2 (38:50):
Yeah, this could be.
I don't know the story, by theway, with the Russian geometry
kernel.
Speaker 1 (38:54):
It's sort of like linked in folklore for me.
Yeah, it's fascinating.
I don't trust everything that'son Wikipedia, but I think the
general information on it isaccurate mostly, but some of the
stuff that's like totally outthere, um, you require some deep
digging, uh, and also trueunderstanding of the space,
(39:17):
which I do not have.
Um, all right.
So, um, in terms of 3d printingI know you're not entirely
focusing on it I would sayprobably it's accurate to say 3d
printing industry benefits morefrom metafold.
The metafold benefits from 3dprinting.
Is that right?
Speaker 2 (39:38):
maybe.
I mean, I think it's just thatas a business we couldn't
sustain ourselves in thatindustry, but we're still very
keenly interested and happy tosupport people.
Speaker 1 (39:45):
Yeah well, the reason I mentioned is that I think I
want people in the 3d, the 3Dprinting industry to understand
some of the unique benefit ofshape analysis, because they
probably have a certain type ofgeometries that would benefit
more than the mesh-basedtraditionally.
Totally yeah.
So do you want to unpack alittle bit of how your geometric
(40:07):
analysis would benefit 3Dprinting?
Speaker 2 (40:11):
Yeah for sure.
So 3D printing I mean, thereason we even started in 3D
printing is because thegeometric constraints are so
open here, right?
So people probably haven'theard you have free complexity.
Well, this is sort of true inthe sense that you can probably
print just about anything, butcan you actually represent it in
(40:33):
CAD?
And then, more relevant to whatMetafold does, can you
understand it before youmanufacture it?
And so I think, yeah.
So maybe this is the right timeto kind of get into lattices and
metamaterials.
Yeah, this is the and this willbe known to many of your
(40:54):
listeners but one of the coolthings about 3D printing is that
we can create lattices andmetamaterials.
So metamaterials are a way ofgetting new and different
material behavior just using onematerial.
New and different materialbehavior just using one material
.
So your three-dipinter isspitting something out and then,
by controlling the geometry andthe way that it packs together,
(41:15):
we can get really interestingand new material behaviors out
of that.
I honestly believe this is sucha huge area of potential, but
we're kind of somehow not readyfor it.
But I'm excited for the daywhen this can be more broadly
embraced.
But coming back to implicitgeometry and metafold one of the
(41:39):
great things about implicitgeometry and signed distance
fields and signed distancefunctions is that we can
represent very complexstructures very efficiently
Because remember, it's just afunction.
So when you have a function,it's very different from having
a surface representation, like atriangle mesh, where you have
(41:59):
to record the details of eachand every triangle that is,
mapping that very complexsurface.
So people are familiar with theheadaches of large file sizes
and processing time and all thatstuff, and this is where
implicit geometry really shines.
You can both represent thatgeometry but, relevant to
Metafold, we can also simulateit and we simulate it in the
(42:24):
same representation.
So there's no need to take yourimplicit representation and
tetrahedralize it and put itinto a solver.
We can do this all in one go.
So so Metafold and implicitgeometry more generally is a
great way to understand thekinds of shapes that you want to
produce with 3D printing.
Speaker 1 (42:46):
Yeah, um, you said that meta material is too early
of a project for you guys totake on.
Kind of now, I kind ofunderstand why you're pulling
out of the design creation partof the CAD process or the design
process, because maybe we justhave too many ideas.
Speaker 2 (43:07):
Yeah, I mean, there's still space for this, for sure,
and there's some good playersalready doing this kind of work,
I think and we continue, by theway, to support people who are
using metamaterials and 3Dprinting space for this, for
sure, and there's some goodplayers already doing this kind
of work um, I think and wecontinue, by the way, to support
people who are usingmetamaterials and 3d printing
like it's, it's still an activearea.
Speaker 1 (43:20):
It's just that it's not like a market yet and is
there any significant differencein computing power between the
mesh basedbased representationand the signed distance field
representation, these twodifferent ways?
Speaker 2 (43:38):
Well, yeah, I mean it's very compact.
The signed distance fieldrepresentation is very compact,
so you can end up with a veryconcise description of shape
that doesn't need like thesehuge file sizes.
You know there's still.
There's always a trade-offbetween accuracy and time.
You know, like with signdistance field, in theory
(44:02):
they're as precise as you wantthem to be, but in practice you
have to choose a representationto evaluate the function at.
So that choice ofrepresentation will depend on
compute, the function at.
So that choice ofrepresentation will depend on
compute.
It's very fast, but as you getmore and more detailed, as you
get more and you want to dothings bigger and bigger, you
will pay a certain time cost.
But when comparing to meshes,it's very performant, is there?
Speaker 1 (44:27):
such a day in the future is possible to figure out
what part of a component useSDF and what part use mesh as a
combined product?
Speaker 2 (44:39):
I think this is a great question and I think there
are some initiatives in thisdirection.
It's not what we've chosen tofocus on, but I think that is
the right question to ask,because there are certain
structures that are very wellrepresented by conventional,
conventional CAD representationsand there's some that are
better in implicit and so how doyou bring them together to find
(45:00):
this kind of like happy, happycombination really?
Speaker 1 (45:05):
and maybe also large amount of existing pre-existing
mesh that's out there that canuse your software to analysis.
Now coming to the more, perhapsa little bit futuristic and
exciting topic, which iseverybody is talking about AI
and machine learning these days,even though these topics really
existed for 40, 50 yearsalready and because there's
(45:29):
money pouring into these names.
Can you talk about how you seethis kind of wave of excitement
behind AI, ml and relevant toMetafold?
Speaker 2 (45:45):
Yeah, it's a good topic.
I mean maybe like the firstthing to say which I find people
are often confused about butMetafold is not an ai platform.
Right, we are just not.
It's not based on largequantity of data.
We're not training models.
We're based on that.
So that's the first thing tounderstand, but I think,
(46:06):
building on that, um, what'sinteresting is is coming back to
again, like our reason forbeing.
We're converting geometry intoinformation and ultimately, that
information should be the inputto any machine learning models.
So we certainly work withcustomers who are building
machine learning models I wastalking about the costing models
(46:27):
, for example.
This is an ML technique and weconvert three-dimensional data
into a feature vector, somethinglike literally a vector of
numbers that a machine learningalgorithm can ingest very easily
.
So the bigger point here, andsomething I've been thinking
about for a really long time, isthat AI and 3D do not go
(46:52):
together very well, and I evenclaim and people will, I think,
probably push back in some waysbut AI for 3D is not there, it's
not happening yet and it mightbe indeed a very long time for
it to happen.
There's a few reasons for this,but if you think about the LLMs
(47:15):
, for instance, the volume ofdata and the simplicity of that
data really is tremendous.
We don't have anything likethat for 3D.
And not only that the data we dohave tends to be proprietary.
There are very few, a verylimited amount of public 3D data
(47:38):
.
So this is one of only one ofthe challenges but, like the
other challenge is that humans,we understand the 3D world so
naturally and so easily.
And think about the last timeyou interacted with an LLM and
it made a comical error that AIdoes not understand anything.
(48:00):
So thinking about how you wouldeven extrapolate that level of
intelligence quote, unquoteintelligence to 3D when we don't
have enough data and it's justa more complex space, it seems
very, very difficult.
So I mean, there are subsets ofcertain applications and
(48:21):
subsets and very targetedproblems where ML can be very
impactful with starting with 3Ddata, but it's a very small
minority of cases.
Speaker 1 (48:32):
I would say it sounds like Metafo basically generated
raw data for some of the dataprojects of starting with 3D
data, but it's like a very smallminority of cases.
I would say it sounds likeMetafo basically generated raw
data for some of the dataprojects of various companies,
but you do not own those data,unfortunately, so you can't
actually create your internalmodel for any kind of future
applications, unfortunately.
Speaker 2 (48:50):
Yeah, I mean for what it's worth.
We are working on this.
We are actually collecting adata set.
One of our current projectsinternal R&D projects is
collecting a data set so that wecan build our own models, just
so that we have.
Yeah, as you say, everythingwe've done is always on
proprietary data, so if we ownour own data, then we can just
talk about the results a littlebit more clearly.
But it's very limited and it'shard to find the data.
(49:14):
It's really surprising maybe notbecause this is a really new
field is that we don't have anykind of open source for the
there's a little bit and there'ssome great data sets that have
been created by the academiccommunity and we love them and
we use them, but it's sort ofnot.
It's not enough if you thinkabout again, the quantity of
data that's feeding these verylarge models that are having
(49:35):
this big transformative impacton everything.
All this AI stuff.
We think about it all the time.
It's something we, as atechnology company, we have to
keep our eye on, but I kind ofalways come back to the same
point, which is that there'sstill a strong need for very
fundamental mathematicalrepresentation and results on 3D
(49:59):
geometry.
That will be the first steptoward any kind of meaningful AI
.
Speaker 1 (50:05):
Yeah, it's like building the foundation of
future AI.
Yes, exactly, and reallyinterestingly, I think any kind
of technologies that aim tointeract with external world,
three-dimensional world, willhave the same problem.
I read a really good articlewhich I can share in.
The link is about how robotics,the field of robotics,
(50:28):
especially the ones that we'reenvisioning, which is autonomous
, intelligent reasoning roboticsthose are not going to happen
either for any time.
So I think autonomous vehicleis probably going to be the
first one, and the major reasonfor that is creating a data set
that's useful for this, and thecost is already enormous
(50:49):
billions of dollars, decades ofefforts into it and it just
simply moved a needle so far.
And I would assume, like us, ourfield 3D printing or 3D
geometry, this kind ofinterfacing with external world.
We have the same data challengeand software challenge,
(51:11):
essentially to really envisionthat autonomous, automated
process.
So, yeah, well, fascinatingworld.
I hope the process isaccelerating instead of a linear
progression, yes, but rather anexponential field.
But the good news is to buildup that foundation, metafold has
(51:33):
tons of business yeah, well,yes, I mean, I sure, I sure hope
so so, um, okay, so we'rereaching the end of the podcast.
I have a couple of fun questionsfor you, okay, um, what are
some of the surprising moments,or many, oh oh, my god, there's
(51:53):
uh, there's been a lot ofsurprises.
Speaker 2 (51:55):
I mean.
Um, gosh, where to even beginwith this?
I mean, I think I thinkhonestly, I've kind of already
said the biggest surprise, whichwas understanding our
technology for analysis overdesign, because we weren't I
don't know somehow this one wasnot on our radar, like and it's
all in kind of the framing too,like our technology didn't
change from having that thought,you know, before and after, but
(52:19):
our understanding of itcompletely changed.
So this was a big surprise.
Um, there's surprising moments,I don't know.
There's more kind of comicalthings, like, in fact, when we,
when we began, we actually builta 3D printer and that's how we
got started as a company.
We were building this reallycool 3D printer and it was so
great and we were talking toinvestors and we had this one
(52:42):
set of investors who was veryinterested and we were building
the software and the hardware atthe same time and they were
very interested and they saidyou know, I think your strength
is in the software.
Like you've done something veryinteresting in software.
We're not as interested in thehardware because there's like a
lot of 3D printers on the marketand me and my two co-founders
(53:04):
went away and you know like ourhearts were in this printer.
You know, you get so attachedto these ideas and so we had
this big soul searching sessionand then we're finally like,
okay, we'll just do the software.
So we kind of got our acttogether on the software.
We went back to the sameinvestors and said like, hey, we
did the thing where you'repivoted, we're just doing
software now.
And they were like, oh, this isso great, this is so great.
(53:24):
We, um, we don't invest insoftware.
So I mean and there's been a lotof a lot of things like this,
you know, just um yeah, it'sit's just a journey and and I
think all the, all the little,the big and small learnings have
have kind of been surprisingalong the way you will think
(53:46):
that investor will tell youbefore.
I know, I know, yeah, it feltlike, okay, this is going to
unlock this next stage, butanyway, we were.
We were pretty nice.
Speaker 1 (53:55):
So many moons away.
I remember you joined one ofour virtual events and I think I
don't know if it's because ourpersonal conversation or maybe
the events you were talking tosomething about life, work,
balance and you're probably oneof the few founders actually
talk about that.
Not to mention, you're also oneof the few female founders who
(54:16):
also uses masks extensively.
So many unique features.
So how do you balance as afounder?
Speaker 2 (54:26):
I guess the first thing to say is I don't know if
I really believe in this idea ofbalance.
Speaker 1 (54:32):
Like if I'm honest about it so it's.
Speaker 2 (54:36):
it's not a balance per se, um, I think it's more
that I've I've learned enoughabout myself to set boundaries
and to know what I need to feellike a human being and uh, so
you know concretely this work wedo.
It's so you know, it's in ourminds, right, it's in our minds.
(55:00):
It's not embodied in any way.
It's, yeah, it's easy to kindof get stuck in that world, and
so I really think a lot about,just like, movement I'm an
absolutely obsessive exerciser,this keeps me sane but also
things like handwriting, youknow, getting off the computer
and and handwriting, um, so thekinesthetic action of, of
(55:24):
writing on a piece of paper witha pen, um, and similarly,
drawing.
So I'm a big believer thatwriting is thinking and drawing
is seeing, and so when I want tothink, I write and when I want
to see which is well, I alwayswant to see, but I just I like
to draw and I think that's animportant practice as well for
(55:45):
me.
Kind of quote, unquote balanceis like, well, drawing
boundaries around my time andthen making sure I have time to
like be a human, a physicalhuman, in the world.
Speaker 1 (55:59):
Yeah, I have to say you're really a unique person as
a combination of artists butalso mathematician, which
basically is a concrete set oflogics, are actually quite
inflexible.
In a way.
You have to say one plus one isequal to two is absolute truth
and several steps of logic goingtowards a certain outcome.
(56:21):
So it's, it's fascinating tosee both features in one.
I have to say my, uh, my brainis malfunctioning, short
circuiting at the moment.
Speaker 2 (56:31):
Well, no, but I, I but I, I mean I might push back
on that a little bit, becauseone of the I'm actually terrible
at arithmetic, like terrible.
Um, that's not the way my brainworks, but I think, uh, for
mathematics, I think somethingpeople miss about mathematics is
that it's very creative, verycreative.
So this is really like thelinking theme between between
(56:56):
what I do in mathematically orartistically or whatever like,
and and also what I do as abusiness person, like I'm.
I'm interested in facilitatingcreativity in our customers to
allow them to achieve thesethese, uh, these creative
engineering outcomes basically.
But I think that is for me whatI've come to as the linking
theme between mathematics andart.
Speaker 1 (57:17):
Yeah, I mean.
Some people said thatmathematics is nature's language
.
Speaker 2 (57:21):
It's universal.
Speaker 1 (57:23):
So well, just for fun , what do you read or observe or
listen to to get inspirations?
Any kind of external mediasource that our listener can
also try to go after?
Speaker 2 (57:43):
Yeah, I mean I try to read broadly, like it's easy to
get into, like the way ourmedia landscape is, it's easy to
kind of get into this um kindof place where you're just
receiving content that's kind oflike tuned to you.
So try and get out of that alittle bit and I read like more.
Let's say, in the past fiveyears, become more and more
(58:05):
interested in, in readinglong-form books, like just
actually books, not long booksbut just books, and the reason
for this is that, like so muchof the information we receive is
very short, like very shortform, and actual ideas, like
real ideas, are messier andcomplex and need maybe a
book-like thing to really getinto.
(58:26):
So I try to read real books.
I read them very slowly, soI'll just put it like this I
haven't read that many books,but I can maybe give one
recommendation of a book I readrecently which again is like
unrelated to anything else, butit's called Underland.
It's a nonfiction by this author, robert McFarlane, and he's a
(58:52):
nature writer but also kind of aphilosopher and a poet.
So it's beautiful.
It reminds you of why theEnglish language is, uh, is
interesting, it's very rich.
He like expands my vocabularywhen I read him.
But Underland is all about ourrelationship with everything
under the surface of the earth.
(59:13):
So it's about I think thesubtitle of the book is called A
Journey Into Deep Time and itputs kind of so much perspective
on our everyday busyness whenyou think about, like, how long
rocks have been there, forinstance.
But it goes much, much deeperthan this.
He looks at all kinds ofunderground exploration and the
(59:35):
idea of burial and just ourrelationship with the surface of
the earth.
It is fascinating and range andbeautiful and uh, and that's
the kind of thing I like to readis something that's like very
different from from my, my, myday to day and uh, and provides
like a different context.
Speaker 1 (59:56):
And a bigger view of the world and life Way bigger.
Yes, exactly, okay, and Ipromise you.
This is a one last question Doyou have any advice for our
young listeners?
Speaker 2 (01:00:10):
Young listeners.
Well, too many.
Speaker 1 (01:00:14):
I don't know from from, uh, I think well, you know
, right now, I have to say, weare in this wave, or somebody
called bubble of ai, and a lotof new graduates actually in
computer science and mathematicscan find jobs.
Speaker 2 (01:00:34):
So this is the rumor, I don't know.
Yeah, so maybe you have ourdata, jenny.
I don't know.
Speaker 1 (01:00:45):
I read something in Wall Street Journal just
recently saying that the lowertier computer science engineers-
hiring is decreasing.
Speaker 2 (01:00:49):
So yeah, yeah, so.
Speaker 1 (01:00:49):
I believe this.
Speaker 2 (01:00:50):
However, I would draw a pretty firm distinction
between that and getting amathematics education.
So I think I guess my advicewould be study mathematics.
Yeah, not necessarily becauseyou're going to become a
(01:01:11):
mathematician, but rather thatit's a.
It's a discipline for how tothink and how to reason through
arguments, and it teaches you acertain way of of approaching
the world that I think isfocused on searching out truth,
which is something I really careabout, and and you know asking
(01:01:31):
questions like asking reallyfundamental questions, that, and
you know asking questions likeasking really fundamental
questions that can yieldinsights.
And so, yeah, study mathematics, become a problem solver, and
these skills are indispensable,no matter where you go.
Speaker 1 (01:01:46):
Absolutely.
I love that advice and,honestly, the only class I
remember thus far is my mathteacher, is my mathematics
classes In a good way, johnny,in a good way, and not just
learning how to do things, buthow to solve problems in many
different ways creatively.
Thank you so much for joiningus today, Lisa.
(01:02:08):
I hope to see you soon again.
Speaker 2 (01:02:10):
Yeah, it's been my pleasure.
Thanks for the invite, jenny.
It's been a great conversation.
See you soon again.
Yeah, it's been my pleasure.
Thanks for the invite.
Speaker 1 (01:02:15):
Jenny, it's been a great conversation.
See you next time.
This podcast is for educationaland informational purposes only
.
The views expressed do notconstitute medical or financial
advice.
The technologies and proceduresdiscussed may not be
commercially available orsuitable for every case.
Always consult with a licensedprofessional.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com