8 Bits with Cassie Breviu!

This week we are joined by Cassie Breviu!

Cassie is a Senior Technical Program Manager working on the AI Frameworks team at Microsoft. She focuses on the product ONNX Runtime, which is a product to deploy models cross-platform and improve inference performance.

Episode Transcript

Brandon Minnick (02:00):
Welcome back, everybody to another amazing
episode of eight pits the showwhere we interview the people
behind the tech to share theirjourney and hopes to inspire
more people to do the same. I'myour host Brandon Minnick with
me as always is my amazing cohost PJ Metz. PJ, how's your
week? I had COVID.

Pj Metz (02:24):
Let's go ahead and open with the truth. We are truth
tellers. No. So II had a weekend to myself. Last
weekend, my wife went to visitsome family and I was like, Oh,
I'm gonna go there's thisoutdoor beer festival and I was
excited about it. I was like,and it's outdoors. And then I
was feeling really brave. So Iwent to an indoor concert the
next day. And guess what,wearing a mask on the edge of

(02:46):
the crowd. The whole time, abrief 15 minutes in the pit
because they're they're a bandthat requires a mosh pit. The
mask came off. I was reallytrusting my vaccines. And I
trusted him a bit too much thisbat is no joke. So caught it got
sick had been isolating. Ihaven't really left this room

(03:07):
that I'm in the office side ofthe room over there's my bed and
it's kind of been just back andforth for the past six, seven
days.

Brandon Minnick (03:18):
That's tough, but you look like you're doing
better. It sounds like thatfeeling better.

Pj Metz (03:24):
drink Gatorade and lots of lots of chicken Lipton
chicken noodle soup extranoodles.

Brandon Minnick (03:31):
That just sounds nice.

Pj Metz (03:32):
I don't listen.

Brandon Minnick (03:33):
I mean, I don't know about every day but I can
get

Pj Metz (03:37):
adjusted doesn't want extra noodles. How are you
doing? Oh, good have COVID Didyou

Brandon Minnick (03:45):
know I was I was also outdoors. This past
weekend. We went camping down inSequoia National Park and Kings
Canyon National Park, which aretwo adjacent national parks down
in Southern California. So itwas amazing. Really cool because
sequoias got the big redwoodtrees. So yeah, it's like oh,

(04:10):
gosh, no, you want a peopleholding hands around, it
probably still wouldn't evenmake it around. Just oh my god,
insane. And so. And then youjuxtapose that with Kings
Canyon, which is this hugecanyon created by glaciers back
in the day. So very much similarvibes to Xi'an if anybody has

(04:30):
been designed on or evenYosemite kind of just you feel
small. You're surrounded bythese giant trees, these
mountains coming up out in whatseems like nowhere made out of
granite. And it's just like, Youknow what? All that stuff. I
thought matters. It doesn'tmatter.

Pj Metz (04:52):
It's so nice to have that place in the universe, like
firmly secured like the firsttime you see the Milky Way and
you're like, oh, like justnothing. Advice From that moment
on, you're like, Yeah, I'm good.
I don't I'm not worried aboutanything.

Brandon Minnick (05:04):
Yeah, so that was lovely. It was our first
time taking a road trip in ourin our Tesla, so figuring out
how to charge it, when you'realso staying at a campsite
that's off the grid wasinteresting.

Pj Metz (05:19):
Fun.

Brandon Minnick (05:21):
Yeah, it basically meant we drove an hour
back into town back into Fresnoevery morning, charged all the
way up to 100%. That was enoughto get us around all day. And
then we do it all over the nextday. So they should probably
average channels on that thing.
We were talking about thatthere. I've seen some mods out
there that people have hackedtogether. But

Pj Metz (05:42):
just imagining like pulling a trailer that's got a
whole like array of like, thetrailer 36 feet long,

Brandon Minnick (05:51):
easy. Now the one I saw, it kind of pulls like
almost like those barbecuetailgates that you can put that
kind of pulled out. It lookslike it reminded me of that,
where you get to the campsite,then you pull this out of your
trunk and kind of unfurl thearray. But we made it, we
survived. I mean, we you have todrive a lot around, especially

(06:12):
these ones that are just so big.
And so I don't even think thatwould have saved us because
those, those guys used a lot ofjuice. But there's a lot of fun.
And big announcement. We arerecording this today on May 24,
which is the first day ofMicrosoft build. It's the big
yearly developer conference thatMicrosoft hosts, which really

(06:33):
means it's the time when we atMicrosoft debut a lot of our
products. So the product thatI've been working on for the
last year of my life is thedotnet Malli community toolkit.
It's finally live. It's been outin preview for months now. But
we've just launched v1 Thismorning, really early this
morning, because Gerald my buddywho releases it lives in the

(06:57):
Netherlands. So as soon as hewoke up, he shipped it. So
beautiful announcement to wakeup to. If you're a dotnet
developer, check out dotnetMaui, which is also being
released today. It's a way tobuild cross platform
applications for mobile anddesktop. And the best part is,
it's all in C Sharp. And they'reall using the native API's

(07:18):
underneath. So totally nativeapp that you can write in C
Sharp running on Mac OS,Windows, iOS, Android, you can
use the dynamic communitytoolkit that I made to make your
life easier. Go check it outtoday. It's It's really amazing.
It's been a lot of a lot of hardwork from the team. So good to
take a moment this this weekendto just breathe.

Pj Metz (07:44):
Finally, I just realized that I missed two days
ago, we had a big thing get livetoo. We just had getting like 15
release on Sunday 22nd of everymonth we released a new version
and we just hit 15. So we're at15 Dotto and we're super stoked
about it. We're gonna actuallyhave like a several month long
celebration. Cool stuffhappening at GitLab as well.

(08:05):
Man, we work for some goodcompanies, huh? Also you have
been working on that toolkit forsuch a long time. I'm so like,
it's alive. You know what Imean?

Brandon Minnick (08:16):
Thanks. Yeah, it's like the first product I've
birthed at Microsoft fromliterally nothing to starting
conversations with lawyers onthe legal team and engineers on
the docs, team and mark

Pj Metz (08:28):
when legal talks. So I'm glad that you had a good
one.

Brandon Minnick (08:33):
But anyways, we have somebody as a guest today
who is way smarter than me, wason a product that's way cooler
than what I do. And I'm soexcited to have her on because
one of the one of the firsttimes we met she was actually
keynoting at a conference. And Iwas like, Who is this person? So
without further ado, Cassie,welcome to the show. Hey, thanks

(08:58):
for having me. Thanks so muchfor joining us today. For the
folks who haven't met you yet.
Who are you? And what do you do?

Cassie Breviu (09:08):
Yeah. So what I do right now, I've had a lot of
different roles with Philipprobably talk about a little
bit. But currently, I'm a seniortechnical program manager on a
product called Onyx runtime,which is part of the AI
frameworks org at Microsoft. Andwhat Onyx runtime does, it's
actually an open source productthat allows you to do machine
learning inferencing on multiplehardware platforms, like fusion

(09:30):
providers languages, so usually,you know, you're building models
in Python or R. And thennecessarily, when you're
deploying and going to use thosemodels, you're not necessarily
using them in Python applets,you might be using a C sharp,
maybe you want to use Maui, likeyou're talking about. So this
would actually allow you, ourproduct allows you to do that.
So it's a runtime that allowsyou to take these models that
you've converted to an onyxformat and make them run

(09:52):
everywhere.

Brandon Minnick (09:54):
That's incredible. Is when it's so
easy. be on the same teamtogether at Microsoft. And like
I mentioned one of the firstconferences we ever attended
together, you were literallykeynoting. Yeah. But you're also
still very new to your career,which was incredible, because

(10:14):
the keynote was amazing. Huge,huge applause at the end of it.
So let's, let's go way back. Welove hearing everybody's origin
stories on apex, becauseeverybody has a different path
into tech in into where theircurrent role is today. So let's
go way back. When did you firstlearn to code? And how did how

(10:36):
did you get into all this?

Cassie Breviu (10:38):
Yeah, so I am one of those non traditional
backgrounds. And it actuallystarted when I was a data
analyst at a company that was athird party administrator for
benefits. And we needed to dothis eligibility audit for
insurance. And it was actuallysupposed to go to the
engineering group, but we hadsome in house developers, but
they were too busy. So it landedon my desk. And so it started

(11:02):
with me having to do this largeaudit manually. And I didn't
want to do that. So I startedlearning, Excel macros, or will
Excel formulas, which thenturned into Excel macros. And so
I clicked record in Excel andrecorded a macro and then viewed
source and then hit play and sawthat I had just created code

(11:23):
because it's VBA. And I waslike, What is this magic. So
that was like, literally wheremy journey started was in Excel,
trying to try to automatesomething I didn't really want
to do.

Pj Metz (11:34):
I've met a lot of people who started out with
like, I just need to automatesome part of my job that
involves data. And so then theystart down that road, and then
they're like, Wait, what's this,and then they like, they get
like, distracted and start awhole new career.

Cassie Breviu (11:49):
And so then once I had like, automated half of my
job, because I did so much inExcel, and so much with
different formulas, I startedgetting macros that would call
other macros and so like to runmy, like half day work, I would
just kick off my process, youknow, debug it if it got into an
issue. And then it was like,Well, I really liked this coding
thing. And like, I think I justwant to do this, like, I think I
want this to be my job insteadof using it to not do the stuff

(12:09):
that I don't want to do in myjob.
And so that's kind of where itstarted. And then once I wanted
to move, I knew that's kind ofwhat I wanted. And I was like,
Well, I kinda want to get intolike a technical role. So I
actually went found a helpdeskrole at a small company that did
non credit, continuingeducation. And from there, I

(12:30):
worked in their helpdesk while Itook classes at a community
college. And so I actually didend up getting an Associate's in
computer science degree. Now,this was before, like boot camps
existed. And if boot campsaround, I've kind of wonder if I
would have ended up doing that.
But I was looking for somethingthat was not a four year degree,
that was something I could doonline. And, I mean, this was
like 10, probably, like 10 yearsago. And so that was like, now

(12:51):
there's so many options, like,it's so much easier now. And so
at the time, it was like, well,this seems like a good way to do
it. So that's why I did it. Andthen, because I had put myself
in a technical role I startedgetting exposed to, to
everything. And because it's asmall company, you know, you're
you can be exposed to a lot morein small companies, because
there's not like the sectioningoff of roles, right, like, so I

(13:12):
could see all of these differentthings. I was working with the
developers, I was working withthe business analyst. And then
eventually, I moved into abusiness analyst role there. And
I started designing productfeatures for our product, and
meeting with our stakeholdersand customers and getting back
those feedback and startedcreating these wireframes. And
so that's what I did until I gotmy first development role.

(13:36):
So I always tell people, well,one of the things that I think
always worked for me was maybetaking something that wasn't
necessarily what I thought wasthe perfect thing for me, but it
moves me in the right direction.
And then that helped me get towhere I want it to be. Or I've
been able to also like change myexisting roles in to make them
what I want them to be. So I'vealways kind of had, I guess,
I've been called scrappy, wantto call it to kind of make

(13:59):
things what I want them to be.
And then when I moved into myfirst development role, speaking
of this scrappy nature of mybackground, I was actually a QA
role. And I was like, Well, Idon't really want to do QA I
wasn't my goal. Their goal wasto build applications, build
software, I wanted to code. Andthey're like, well, so after my
first interview with them, and Itold them in my first interview,

(14:21):
they changed my second interviewand made it a developer
interview. And so they actuallyinterviewed me to become a
developer. And granted, youknow, I was brand new, so it's
not like I was amazing, but Icould do some things. And they
ended up hiring me with theassumption that, you know, they
could mentor me into becoming adeveloper. And that's exactly
what happened. I worked in QAfor a few months and then after

(14:43):
that few months, I moved into ajunior development role at that
company.

Brandon Minnick (14:46):
Absolutely incredible. I love is one of the
things I always say is a lazyengineers, a good engineer,
because we look at thesescenarios like, Oh, I got to
spend half a day entering stuffinto Excel. I don't want to do
that. So let me let me justwrite some code, get rid of
that. And so for anybody that'slistening, that has a really

(15:10):
mundane, boring tasks that youhave to do, because we have,
every job has that no matter howamazing your job is, I've
learned, there's always going tobe some mundane, say
administrative stuff you got todo. See if you can automate it.
Because next thing, you know,you might be senior technical
program manager at Microsoft isright, I think what happened.

(15:32):
That's the next

Pj Metz (15:33):
part, actually.

Cassie Breviu (15:37):
Eventually, I always say so the best way that
I learned is like, I have agoal. And I just don't stop
until I figure it out. So like,it was the same thing with Excel
formulas and figuring out thesemacros, like I was looking
everything up. And I was didn'treally know what I was doing. I
was copying pasting things inand trying to figure out what a
class was, and all thisdifferent stuff. And so it's
definitely that, that goalfocused learning is is how I

(16:00):
like to do it, because theneventually get something to
work. And then I peel apart thelayers and I figure out how it
works. And and then when you'redone, when you actually reach
the goal that you've worked sohard for and probably wanted to
give up many times, you realizethat you've learned a new skill.
And that's actually how I gotinto AI too, because now I focus
on AI, which back then I wasdoing C sharp, full stack
development and web developmentand stuff like that. And it was

(16:22):
it was the same thing, whichactually came from being lazy as
well. Didn't want to assignhelpdesk tickets anymore. So
that was a machine learningmodel that I built.

Pj Metz (16:34):
And that's got, you've got a lot of fans of your desk
as a hammertoe on Twitch said, Iactually built my desk back in
the UK based on her blog postsof her desk setup, using steel
pipe to make a frame formonitors, lights, cameras, etc.
So like, not only are you doingexactly what you want, and like
making these great strides inyour job and all this stuff, but
you're inspiring other people tomake better desks. So we applaud

(16:58):
you. That's fantastic.

Cassie Breviu (17:00):
That's awesome.
Yeah, that block has gotten alot of that was actually from
build, oh my gosh, how manyyears ago? I want to say two
years, two to three years ago, Ihad posted my my setup. And like
I got a ton of questions aboutit. So I wrote that blog really
quick, because I had so manyquestions in the chat. I
couldn't keep up with it. And soI was like, Well, I'm just gonna

(17:21):
throw together a blog, and sopeople can see it. So I love to
hear that somebody actuallyactually did it. Yeah, it was
really nice. And the wholereason I built it was because I
couldn't find a mount thatsupported all of the things that
I wanted to mount to my desk.
And so I took galvanized pipeand made my own frame so that I
could sound like a lot of stuff.
If you go on my Twitter andlook, you'll see like I have so

(17:43):
much tech on my desk that aretelling

Pj Metz (17:47):
us before the show that you were like yeah, I just
always want more and more andmore and more. Oh yeah, for my
computer,

Cassie Breviu (17:55):
the gadgets, I have all kinds of gadgets like I
won't sit and list them off, butlike it even things like I don't
even need like I have thismotion capture I'll show you one
of these iPads like Leap Motioncapture control. And essentially
it has different cameras andsensors that allows you to do
all kinds of cool things but oneof them is to control things
with your hands so you can feellike a magician so I didn't set

(18:21):
it up yet but I'm like I reallywant to set it up because
there's an app that you can addthat will actually allow you to
like control your monitor likewith your hands like can do
things like you probablywouldn't do every like Minority

Pj Metz (18:31):
Report like that's what I was thinking

Brandon Minnick (18:35):
like move windows around Yeah, yeah, zoom
in.

Pj Metz (18:40):
I just I just want it so I can do MS Paint and like do
it in the air and like it likethe spray can and like I'm

Cassie Breviu (18:50):
actually I got it because I have this other gadget
on my desk. Yeah, yes. LookingGlass. 3d. It's a holographic
display

Pj Metz (19:03):
that looks like a 3d model. So like she held up like
a picture frame for peoplelistening to podcast and like
there's a picture inside of itbut it looks 3d Like it's got
shadows behind it. But it's justa picture frame.

Brandon Minnick (19:17):
Yeah and as you rotate

Cassie Breviu (19:19):
like it things move in it as well. What Yeah,

Brandon Minnick (19:22):
wow, this is incredible. So yeah, again it's
it almost looks like there'ssomebody inside of a shadow box
moving. But what what is it isit flat? Like is there actually
depth to that and real

Cassie Breviu (19:37):
depth so it is tricking you it's tricking you
into seeing and hollow ahologram but it is a it's a
holographic display and it lookslike I know you can kind of see
it here and they can't see itbut it looks like the future to
me like most what times when youthink about I'm really into XR
and VR and and all of that andAR and all that stuff too. And
so really like most of the waysthat we interact with virtual

(19:57):
like 3d objects right now arereally through wearing something
Right, like you've put on. Thisis the first device that I've
seen that I think does a reallywhat good job and really like, I
mean, it's still expensive, butmore it's more accessible than I
mean the same as like an Oculusquest, you know, that allows you
to do 3d out that withoutputting anything on your face,
like with being able to do that.
And I've seen some really coolkind of like, projects with it

(20:19):
that are actually like doingvideo calls, like where you can
do video calls. In 3d. Yeah. SoI really think this is the
beginning of more of more 3ddevices, at least I hope
because. But yeah, there's,there's so many things you can
do with it. So I got that in theMagic Leap, because I'm going to
build um, you can also buildthings where you can control the

(20:43):
3d monitor with your hands. Solike you'd have that 3d display.
Leave, just like Ironman.

Pj Metz (20:54):
My whole life is lived through movie references, and
we're getting closer and closerto the Riddler is invention from
Batman Forever. Or Jim Carreywas the Riddler where you watch
TV and it's like, 3d. Yeah, I'mjust, it's all superhero movies
in my head, I'm sorry.

Brandon Minnick (21:12):
deep cuts. It is.

Cassie Breviu (21:13):
And if you think about it, I think Minority
Report was like the first time Iremember seeing a truly like
holographic display that youcould move with your hands like,
and I always think it's like isart imitating life or life
imitating art because you seethings in movies that like don't
exist, but they're soimaginative and so cool. And
because of movie magic, they canmake them and I feel like,
eventually some smart personlike comes along and figures out

(21:35):
how to like make it somewhatreal. But it takes time for the
technology to catch up. But Ifeel like sometimes just like
the people that dream up in themovies are the things that then
we try to attain it in tech,like the techniques, look at it
and try to mimic it in ways thatcan be fun, real.

Pj Metz (21:52):
I fully believe in that pipeline. Like when you think
about like Star Trek in the 60sand the gadgets that they had,
what happens is people go, Well,what's keeping us from doing
that? And what's cute, and sothey try it and then they find
great new ways to create techfor us. Oh,

Cassie Breviu (22:07):
my iPad, there's like, do you can find like,
really? blogs that are like StarTrek invented all of this stuff?

Brandon Minnick (22:14):
I was just thinking that yeah, the cell
phone. Yeah, we haven't gottenthe teleporter yet. That's
annoying. Yeah, come on. Thankyou so much. I think this
weekend.

Pj Metz (22:23):
Why can't I teleport to a national park? That's amazing.
Oh,

Brandon Minnick (22:30):
goodness. Yeah.
It's it's interesting, because Iremember back in the day, there
was a there's a while where 3dTVs were like the new hotness.
And I remember that was one ofthe first because I was always a
big technology guy growing up. Ialways buy latest gadgets and
whatnot. Yeah, I just bought aTV before that wave. So I just
started my first job ever out ofcollege, use my signing bonus to

(22:54):
buy a TV blew it allimmediately, of course. And and
then yeah, 3d TVs came out waslike what? Like, I just spent
all this money on a new TV.
Like, just out of spite, I hopethis doesn't become a thing,
because then I gotta buy anothernew TV. And thankfully, it did

(23:15):
well for me. So I didn't have tobuy a new TV. But yeah, I I
always said because I didn'tenjoy wearing the glasses, like
the 3d TVs you have to put onthe special glasses sync up to
the frame rate. And it makessense. But I was like, Wouldn't
it be cool if you could have aTV? That just was 3d. And that's

(23:35):
what this reminded me of. That'swhat that picture frame reminded
me of his. We're just looking atit right now. We'd have weird
glasses on or anything special.
And yeah, it's 3d. So

Cassie Breviu (23:45):
they do make a big one. It's just insanely
expensive. I think it's like$3,000 or something like, but
there is actually like, I thinkthree different sizes. I have
like the, like the smallest one.
But yeah, they're relatively Ithink newer company. And they do
have a large one. And I got touse it at the mixer or an MIT
reality hat. They had a largeone. And they had like a, like a

(24:06):
spaceship thing. And you couldpull it apart with your hands.
Like without actually touchinganything, and you could like
move it around and play with it.
So I always wondered why thosethe 3d TVs didn't take off. But
if you think about there's somany things that have been
created and failed, but thencame back better, like even VR

(24:29):
has failed multiple times.

Pj Metz (24:32):
We all remember that.

Brandon Minnick (24:34):
But this time

Cassie Breviu (24:40):
you know, it takes time for things to catch
up for the technology, thehardware, really to catch up for
what we can dream up.

Pj Metz (24:46):
The only thing I thought the 3d TVs were really
really good at was you could setit up to where multiple people
could be playing a video game onit at the same time. And they
would just put all the imagesover each other and Everyone
wear glasses and everyone got afull screen view. It was super
cool. And it's literally a yearafter that everyone was just

(25:09):
playing online instead. playmultiplayer in person anymore.
But you can't screen watchanymore because it's just your
screen. And then finally I'dmaybe be able to beat my
freshman year roommate at Halo.

Brandon Minnick (25:24):
Right back, back in my day, we had to play
Goldeneye one TV.

Pj Metz (25:28):
Yeah, it was a four by three 300.

Brandon Minnick (25:35):
And then you chop that up into four screens
for four players. Yeah, youdidn't get a lot of pixels back
in the day had

Pj Metz (25:39):
the tape cardboard on the line. So.

Brandon Minnick (25:45):
Oh, goodness, so. So Cassie, here we are in
your journey. So we are nowworking in QA working towards
this developer job. What happensnext?

Cassie Breviu (25:59):
Yeah, so once I finally got the role, like the
engineering role, I was soexcited. But I was also like
crazy imposter syndrome. And Ididn't even know that's what it
was called back then. But Idefinitely had that I was on an
all male dev team, that I wasactually it was a law firm that
I worked at an intellectualproperty law firm. And many of
them had been in tech for a verylong time, most of them were

(26:21):
senior level. So I was superintimidated as well. Like, I
just remember my first stand up,and I probably like a deer in
the headlights. So I was like,Oh my God, how did I get here?
Well, and then, and theneventually, after a while, you
know, at all, it was fun. Andthey I had some great mentor
there and all of that. But so Istarted learning, I started

(26:44):
building different applications.
And the thing that was reallycool about there is we built
stuff from like, I would get inlike a business, I was building
apps for like internal lines ofbusiness. And so they would need
something to solve a problemthat they had to do manually.
And so then I would have to goand kind of like I would design
it. And then I would build uplike the database, I build out
all the. So I did like all SQLstuff, which was and then I did,

(27:05):
you know, the C sharp back end.
And then I did the front endwith JavaScript. And like, I was
still doing the whole thing. Andthen like demoing it to the the
users in the company to see ifit would work and what they need
to change and all of that. Sothe thing that was really,
really cool about that is a lotof times when you go into tech,
and particularly now there'sreally a lot more like

(27:26):
specialized roles. And there Iwas getting this like broad
scope of everything. So I wasvery much immersed and was able
to learn so much. We're allhosted on prem then as well,
because cloud wasn't what it istoday. Now, either. And also law
firms had been slow, in generalto pick up cloud, but so I just
kind of kept going. And I keptbuilding things and, you know,

(27:49):
breaking things on the way aswell. Because of course, there's
always like the legacyapplications that you have to
kind of like reverse engineerand figure out. So we got to
work with a lot of differentlanguages because of that, as
well. So I still think I gotlike one of the best first
roles, at least at least forsome free people that learn like
me who like learned by doing,because I just got thrown into
like all of this information.
And I got to learn so much aboutthe full process. And and yeah,

(28:12):
so that was really cool. Andthen one of the things so I was
in that and I wanted to learnartificial intelligence, like it
was something I had, like seen,right, like, and I was like,
What is this cool thing? Like?
And I just thought it seemed Ididn't know where to start. And
I thought it seemed Oh, it'smust be too complicated. And
what happened is I went to aconference talk at that

(28:35):
conference, actually the onethat you saw me keynote at. And
this was before I was keynoting.
This was years before that. AndI went to a talk and they went
did a talk on how to survive theTitanic, or would you survive
the Titanic? Sorry, not howwould you survive? It's in a
dataset. And they went throughthe full machine learning model

(28:59):
model building process throughthis thing that's now called
designer, Azure Machine LearningDesigner. And I was like, I get
it. And like, I mean, not like Iunderstand all the machine
learning and not what wasn'tlike Neo and like The Matrix.
Forward is really what happens,and an understanding of what the

(29:19):
data science process was. And soas I said, we're talking about
earlier how lazy we had toassign these helpdesk tickets.
And what would happen is, so wehad a help desk there, but they
didn't know because we had somany different in house built
products. The helpdesk didn'treally know what would go where
and they couldn't necessarilytell from the ticket. So it
would go into this like binof A, that would then we'd have

(29:41):
to sift through and figure outand nobody wanted to do it. So
the developers would have tospend a week like maybe I'm
every month where they wouldjust have to have that
distraction like if somethingcame up because because with a
tickets like this, they have tobe triaged quickly, you count
them sit, so you'd have to breakyour workflow and as you know
with you In engineering and withbuilding things like you can get
very into things. And it's hardto once you break that

(30:04):
concentration, sometimes reallyhard to get back in. So it was a
very kind of, like, frustratingthing to have to do. And so I
was like, Well, I think I coulduse machine learning to take the
text of this ticket and figureout who it should be assigned
to. So I don't have to do itanymore. And so that's actually
the first model that I built,was how was assigning helped us
tickets. And, and the way that Idid it is we had our helpdesk

(30:27):
ticket was ticket system, wehosted the data in house, but
the software was purchased, Icouldn't actually integrate into
the code, I could only use thedatabase in order to create
events to do machine learning.
So I, I had built this in thecloud now. So I've used this
drag and drop. So if you're notfamiliar with Azure Machine

(30:48):
Learning Designer, it's like adrag and drop visual tool that
allows you to like build machinelearning models, without writing
really much code, you can writecode, but you don't necessarily
have to. And so I had hooked upto the database and had a
trigger that would kick off theworkflow if there was a ticket
that was assigned to this bucketthat we had to triage. And then

(31:08):
I would take the text of theticket, and it would influence
on who it should be assigned toit would take the top three, and
then it would email them post itto a team's channel. And it
would say sincerely, the machineoverload Lord. I didn't really
tell anyone I was building this,like I had, I've kind of like
started, I was like, hey, hey, Ineed access to this database.
It's like why I was like, I justwanted to do something like I

(31:29):
wasn't really telling peoplewhat I was doing, because for
some reason, I didn't, you know,I didn't know if I was gonna be
able to do it. And I didn't wantpeople to get like excited
because if they don't, what ifit doesn't, right, all that
stuff. But people startedgetting help just to get signed
signing. The first model that Ibuilt,

Brandon Minnick (31:46):
like all of a sudden Cassius productivity just
went through the roof, andyou're deciding tickets within
seconds of them being opened.

Cassie Breviu (31:54):
Yeah, it's great.
Like AI was no different. Forme, it was just the same way
that I had done everything else.
I had got a goal figured outsomething I wanted to build,
broke it down into smallerpieces. researched a lot. But
then, so then what happened?
Because now you realize, like, Ibuilt a model, but I haven't
used Python yet. Right? Like Ihaven't actually in at this

(32:14):
point. I haven't even writtenany Python code yet. Like in my
development career, like I hadnot done Python. It was all C
sharp. I still love C sharp. ButI guess TypeScript makes
JavaScript bearable. So I'll saywell,

Pj Metz (32:31):
come here for your hot languages, y'all.

Cassie Breviu (32:39):
Where was I? Oh, right. So then they're like,
Well, this is really cool. Butwe want we want you to do other
stuff with machine learning.
Like they were actually reallyexcited and really supportive,
which I think was one of thegreat things about that place is
they were always like, open tonew ideas, and trying new
things, which isn't always thecase at companies. And so that
was one of the things there thatwas really, really cool is like,
they would be like, You knowwhat, I think that this is
useful. And I could see value,like, why don't you take some
time and figure something out.

(33:01):
So they wanted me to be able tobuild it on prem? Well, this
tool was on the cloud. So I waslike, I only know the models
with this cloud tool. Andthey're like, well take a couple
weeks and see if you can figureit out. And I was like, Okay,

Pj Metz (33:16):
just a few weeks.
That's all you need.

Cassie Breviu (33:20):
Yeah, I know, it's like, you can get something
to work in two weeks. I didn'tunderstand all of machine
learning in two weeks. But so Ireverse engineered what I built
in the cloud, doing watchingsome different stuff on
Coursera, I watched the suppliedAI course on Coursera. And I
kind of just took differentpieces of different things to

(33:41):
figure out what I needed. Andthen put it all together and
then looked at how I had builtit in the tool and reverse
engineer that and I ended upusing scikit learn so I wasn't
doing deep learning and I wasdoing classical machine
learning. And yeah, I ended upre building it. But I mean, I
put in a lot of hours, those twoweeks because I was super
determined. And I was like, Iwant this to work. So I was like

(34:02):
really excited about it as well.
You know, sometimes when you'relike building things, there's
like a push or a pull, right?
Like you have to like pushyourself to do things you don't
want to do. But when you reallywant to do something you like it
pulls you to the point where youcan't stop until you done at
least that's how for me so thatwas like one of those things I
was putting in like a lot ofhours just because I was
obsessed with getting it towork. And I did eventually I did

(34:22):
get it to work. So that's

Pj Metz (34:25):
at this point, Cassie that like if you get interested
in anything given enough time,you will you will be able to do
whatever it is you want. You areintimidatingly smart, and I am
like terrified that like I'm onthis like it's a virtual stage
but like I'm scared that justyou're really cool. Like this is
amazing. Don't mind me. I'mfangirling here. I'm very

(34:49):
excited to have every story herelike so like I just started
researching and like when Istart researching, I get
distracted and then suddenly I'mwatching like the new Thor
trailer over and over againinstead of actually. So like,
you're just what I think you'reshowing is like, if you really
stick with that, first off, thatpush and pull example, is

(35:11):
perfect. Because there are timesyou have to push. And you have
to say, I just need to figureout the next thing, and I just
gotta get through this. Andthere times you can't think
about anything else. And usingboth of those ways to your
advantage is really important.

Cassie Breviu (35:24):
Yes, totally. And some like this was really hard
to like, I'm not I don't, Idon't think I'm extremely smart.
I think I'm just extremelydetermined. And I don't give up
no matter like how many timesand I know that sounds like
really cheesy, but that honestlyis like, I will literally code
till my eyes hurt and like ithurts to blink because I've been
staring at my screen for solong. Like when I do something

(35:45):
like this. And I actually wentout with a friend like that
Friday, like, I think it was thefirst week when I was like
learning everything. Andliterally, I was so tired, that
I like could hardly formsentences. And I just told her,
I was like, You know what, I'msorry. Like, it's gonna be hard
for me to find the right words,because like, my brain is so
tired right now. Because I'vebeen so like, focused on this

(36:06):
thing. But like to talk aboutyour distractions comment, like,
I get like that, too. I justdepends, like, I have to really,
really like want something andthen I don't get distracted. But
like, I have a harder time whenit's something that I have to
like make myself to then it's alot harder to do. So like it was
really hard. Like, you can lookback at it now and be like, Oh,
I did this thing. And it's like,but I worked insanely hard. And

(36:28):
it was like, really peereddetermination. That's really all
like, if you're determined ityou can. And you will.

Brandon Minnick (36:37):
Yeah, that's that's one thing I love about
software, was actually chattingwith a guy at the campsite this
weekend, who used to be asoftware engineer, and now he
runs a campsite. Which is notuncommon for a lot of folks in
tech to eventually retire fromtech, like, retire retire from
tech. And yeah, he gave me ahard time. He's like, Oh, how

(37:01):
long have you been doing this?
Like 1015 years? Like, okay, soyou're not there yet? Yeah,
because I was like, I love towrite code. And one of the
reasons I love it is building onwhat Cassie said is you you can
do anything like there arelimitations. You know, there's,
there's physics involvedeventually. And you can't do
anything anything like that'sprobably why we don't have

(37:25):
teleportation yet. But you canessentially build anything you
want. Writing code. And there's,I've, I found a similar thread
in my career to Cassie, that,there's always a way to do
something and rarely can justnot be done. And so yeah, if you
just keep trying stay focus thatyou'll eventually figure it out.

(37:50):
It might not be easy. But thatis one of the favorite things
that I found when writing codeis eventually I'll be able to
fix this bug, eventually I'll beable to implement this feature
because there's, there's got tobe a way to do it. I just need
to figure it out. And that'salso I found leaked into the
real world, like something inthe house breaks, like I had to

(38:14):
fix the hot water heater acouple of weeks ago, because all
sudden, we ever taken coldshowers. And that's not fun. And
that's because this thing isdesigned to work in a specific
way. If I can just figure thatout, then I can figure out
what's broken. And then I canhopefully by that replacement
part. And so I love that advice.

(38:36):
Just in general, there's there'sa way to do it. If you just
stick with it. And kind of Yeah,break it into smaller parts.
towards that goal. Yeah.

Pj Metz (38:50):
Yeah, it's not it's not fixed the water heater, it's
okay, find where it's stopping,and then look online, similar
and then find that replacementpart like an ace or Lowe's or
whatever, and then figure outhow to like there's steps in
there and knowing what thosesmaller steps are, makes the big
task manageable. So yeah, Cassiedidn't go, well. I'm gonna make

(39:11):
a I was like, oh, I want toautomate a thing. Okay, how do I
do that? Well, I saw this, thispresentation at a conference.
And I took that home and Istarted working with it, and I
started finding things. So it'snever, you don't accomplish the
big thing without the 1000 stepsleading up to it. And speaking
of 1000 steps, here's the firststep towards us getting

(39:32):
sponsored on this show. Ifyou're hearing my voice, that
means you've been listening toor watching eight bits with
Brandon MPJ. And we're here totalk to you about your product
and how it can help you in yourlife by to do whatever your

(39:53):
product does. So if you're anavid listener of the show, or
you watch us on Twitch Then youwill know that your product,
your product is right for you.

Brandon Minnick (40:13):
That's right.
We are just a humble podcastthat still has to pay the bills.
So if you have a product thatyou'd like to, for us to feature
here on eight bits, send us anemail Hello at eight bits.tv.
And we'll be in touch. Now,Cassie, where we left off.
You're, you're you're changingthe world, you're doing anything

(40:35):
you want, you're essentiallyseeing the matrix and making it
bend to your will. Butselfishly, I know, I've known
you for four or five years now,specifically, because we've
worked together Microsoft, andwe haven't even gotten to that
part in your story yet. So wheredo we go from here? What's next?

Cassie Breviu (40:59):
You mean? Like, how did I get from there to
here? Or where I'm going now?
Because I never have any idea.
By the way.

Pj Metz (41:07):
How did you get to Microsoft? Your memory and

Brandon Minnick (41:15):
washing, I woke up?

Cassie Breviu (41:19):
You know, I was like, I like AI. And they're
like, We like aI too. And I waslike, Cool. Let's do it
together, then. Yeah, no. No. Sothat's, that is one thing that I
think is funny. Looking backnow, like, I can say things
very, like, it seems sostreamlined. And it makes so
much sense. And, and but it'sthe hindsight 2020 thing for
sure. Because like, I did notknow, I just knew that I wanted

(41:42):
to do like, I just like when itstarted with the development
thing. I'm like, I want to dosomething with development. So I
got a job and, you know, helpdesk and then it's like, well, I
know I wanted to do this. But Iactually really liked the
business, the design side of it,like when I was doing that
product, feature upgrades andstuff like that, like I really
liked this. And I almost changedinto doing more of just that
instead of moving into anengineering role. But I was
like, No, I'm going to stickwith my original plan or what I

(42:04):
really wanted. But yeah, Ireally feel like it's one of
those there's this like FionaApple song about the roll just
rolls out in front of me. Andthat's kind of how I feel about
it. It's, it's like, I know, Ihave ideas of things that I
want. And it's not that I'm notgoal setting, because obviously
I set goals, and I've createdthings that were allowed me to
get new skills that allowed meto progress, my career. But I

(42:26):
didn't start out, you know, 10years ago, or whenever it was,
like, I'm gonna be, I'm gonnawork at Microsoft, I never, ever
in a million years thought Iwould have ended up here, I feel
so lucky to have ended up here.
But it was just and so honestlyto the point because I don't
have a traditional background.
Right. So talking about how Igot here, I never actually
really applied to big companies,because it was a lot of times in

(42:48):
the HR process, when you don'thave a bachelor's degree, like
you're just they just you neverget called back. And, and even
like with the small companiesthat I worked at, a lot of times
I go through recruiters, becauserecruiters go straight to the
hiring manager, and it skips HR.
And so this gets HR, they're nothiring managers in tech know
that it doesn't matter if youhave a degree or not. And so
that was like one of the waysthat I had to kind of like work

(43:09):
around the fact that I didn'thave a traditional education.
And then I would do temp, Iwould do temp to hire because
then they can try it. Try youout, essentially and see if you
work with it. So almosteverything I did before,
Microsoft was tempt to hirethrough a contracting agency,
because I had to go around HR.
So I never actually would haveapplied to Microsoft to be

(43:31):
honest, but not because Iwouldn't want to I just would
assumed that they would neverhave talked to me. And then I
was at that conference that Iended up talking to some
Microsoft people the year afterI had wanted that talk and I was
having issues with hosting stuffin Azure, with just specifically
what I was doing. I was doingokay, so now the funny thing is

(43:52):
now is like, I looked back and Iwas doing ml ops, but no lapse
didn't really exist yet. I wasworking with operationalizing
models in ways for a startupapplication that I had been
working on. And I was runninginto a lot of issues. And now
looking back, I'm like, well,that's why I was having issues.
It's because there wasn't a lotof people doing that quite yet.
At least not at like in smallcompanies. Right. So anyways,

(44:13):
they asked if I would beinterested in interviewing, and
I was like, Yes.

Pj Metz (44:18):
Really? Talking to me, or is there another Cassie?
Yeah.

Cassie Breviu (44:27):
Yeah, so that's that's how I ended up here. And,
and I'm eternally grateful forthe path that I've had. And I
think a lot of it just came downto putting in the work. And but
I know some people are much morelike planned and they're like,
they have all these goals andlike vision boards and stuff
like that. And I think that'samazing, too. And I think you
get amazing results that way. Ithink everybody's just kind of

(44:47):
different and there's not likeone right way. There's not one
right path. It's everybody kindof has to find their own way.

Pj Metz (44:53):
I guess. We have three totally different paths on the
show. Brandon who always talksabout how he went to traditional
route But like he like yourdegree, Brandon was computer
engineering, computer sciencehardware, computer hardware.

Brandon Minnick (45:09):
So designing microprocessors in college.
Exactly. So

Pj Metz (45:13):
yeah, so not a sophomore. So even your
traditional path isn't astraditional as some people who
literally major in being adeveloper, you know what I mean?
So like, every path is going tobe different. And Cassie, I love
that you said, like, everyone'sgot a different way to get
there, it's just about whatworks best for you another
person's way might not be theway you do it. And so we can't

(45:35):
sort of, I think the idea ofexposing people to multiple
paths is really important. Andthat's why I love hosting the
show. Because I mean, I get tomeet so many people, and every
single person's path is totallydifferent. You got to trust it,
and you got to put in the work,it's not just gonna happen to
you, you got to put in the work.
And, you know, having a degreeis not the only way to do it. So

(45:57):
don't feel like you have to geta college degree to make this
happen.

Cassie Breviu (46:02):
Right, you do.
But you do have to work hardeither way. And I feel like one
of the things I hear so muchfrom people, and you mentioned
something about it, like, youknow, it's not easy that
everyone so many people, they'relike, I want to learn to code,
like, how did you do what youdid did it and I'll tell them,
like how I did it. And there's,I will say a large percentage of
those people don't necessarilyfollow through. And that's one
of the big things is it's youhave to understand like it is
not going to be easy, and it'snot going to be straightforward,

(46:24):
it is going to be really hard.
And that's what makes it great.
Because also like if you thinkabout like development, like PJ,
now that you've been in adevelopment role for a little
while, like you know that youget really stuck on things, even
after even if you've been doingdevelopment, you still get stuck
on things, and you still have toput that hard work in. Stopping
hard, that's the other thingthat I feel like people should

(46:47):
realize is like, if you don'thave the hard work to put in to
get there or to like learn theskill, then you're and if you
don't like that, then you mightnot like the coding side either.
Because it's very much like,yeah, you have to have, you have
to work hard to do it.

Pj Metz (47:04):
And not only that, you don't have to be able to code in
order to work in tech. There's alot of great opportunities in
tech that don't require coding,even my job. It's not I don't
code day to day, like I'm makingtutorials for like students, and
I'm working with education andlike, my I don't write stuff for
GitLab. But I need to know alittle bit for my job. But I

(47:26):
know people that work in GitLab,that don't code at all. And it's
not just people who work in likelegal or finance or department
require code. There areopportunities. And there are
things out there that you haveno idea what the job title is.
But it might be an opportunityfor you. If you had been like
PJ, you're going to be aneducation evangelist, I would
have been like, that's made up.
That's not a career there. Andthere are other people and like

(47:49):
you couldn't convince meotherwise. But like, here I am.
It's, that's that's one of thethings that I try and impress on
my students that I talked to allthe time, it's like, don't worry
about being a developer. Like ifyou want to work in tech, get
some skills keep working. Andit's like what Brandon said
earlier? And what you saidearlier as well, Cassie, it's
like when you come upon aproblem, work to solve it. And
it's sometimes you go around it,sometimes you go over it,

(48:12):
sometimes you go under theproblem. But solving the
problem, there's a millionroutes to do.

Cassie Breviu (48:19):
You know, and I would like to say to now I see a
lot more job postings that don'tsay degree required. And so if I
look back, I think that I wouldhave been less self conscious
about the fact that I didn'thave the degree and trust in the
fact that I had the rightexperience. And so like, that's
one of those things where like,I was probably cutting myself
short because I had like thisinsecurity or like imposter
syndrome about having the rightbackground. And so I think

(48:41):
that's one thing too, that Ifeel like people should realize
is that like you should applyfor the job that you want, even
if you don't think you have theright background, or even if you
don't have a degree, and alsolook for those job postings.
Like a lot of them aren't sayingthey require degrees anymore,
because they're realizing peoplecan can learn these things other
ways. So I just want to makesure that what I said didn't
discourage anyone from like,applying for particular roles

(49:04):
thinking, Oh, I have to getrecruited. It's like no, I just
that was the thing that showedme that I apparently didn't
necessarily matter that I Ididn't have a degree for that
role. So I think it's animportant thing to think about
and to understand.

Brandon Minnick (49:16):
Yeah, let's seeing like you said, Cassie,
the, especially the big techcompanies like Microsoft,
Google, they're all droppingthat bachelor's degree
requirement now, which is sogood, so freeing will bring in,
it's just gonna get more talentand more people from different
backgrounds, which makeseverything so much better. Yeah,
but, Cassie, we only have 10minutes left. I don't know how I

(49:39):
don't know where the other 50minutes went. But you are doing
some amazing things right nowthat I want to make sure we show
off. So where should we start?
Okay, you've got a new productcoming out.

Cassie Breviu (49:51):
Um, I've been working on I'm like, which one
do I start with? Huge, obviouslyon it. So Onyx runtime is a
product that we work on. And Ido all of the Onyx runtime
YouTube channel of videos andcontent there. And so there's a
lot of different things there.
And I post about every two weeksat least I try to post every two
weeks. And I have one coming uplike in this week that I am

(50:14):
hoping to get ready and postaround something really cool
that Onyx runtime data. So we'vepartnered with Epic Games, to
create a neural networkinference plugin in Unreal
Engine five. And so what thisplugin actually does is it
allows you to run neuralnetworks or deep learning models
within your unreal games. And sothere's like a lot of different

(50:37):
ways where that's starting tochange everything, it's still
kind of new, a newer area in theway where you'll see like the
big companies like I don't knowif you saw the one that NVIDIA
did with GTA for photorealism.
Or did you see that one?

Pj Metz (50:54):
That's just I'm just thinking about all the Unreal
Engine five stuff that I've seenlately. And I'm convinced the
uncanny valley is about to getshattered, like I can't tell,
but I saw one it was like of aJapanese train station. And I
felt like I was in Japan. I was,there was amazing. Absolutely,

Brandon Minnick (51:12):
let's just call it real engine. Now. They've
just really made they madereality. It's incredible.

Cassie Breviu (51:19):
They did. Oh, surprising. When you said
something about how like,there's still limitation with
programming because of physics.
Well, in virtual reality, youget to control the physics. And
so anything is possible invirtual, right, you literally
can create anything that yourbrain can think of, which is why
I love XR and VR so much. Andone of the cool things about
applying AI into gaming, is thatit opens up kind of this new

(51:41):
door, and it's very experimentalright now. But I feel like we're
gonna see a lot of really coolleaps and bounds in that area.
And so that's what I'm reallyexcited about for the the new
plugin because it makes it soyou can use Onyx runtime as your
inference engine. And then ithas like open CV is a helper,
which is another experimentalplugin that allows you to do

(52:02):
different computer vision taskswithin it without having to
create the plugins and integratethe open source libraries
yourself into the Unreal Engine.
So that's something that I'mworking on, I'm really excited
about and the the thing withinit. So like, as I was saying,
you know, like there's like thephoto realism, one of the ones
that I did that you'll be seeingsoon is the style transfer. So

(52:27):
are you familiar with styletransfer, like aI models? So I
bet I bet you've seen it, whereyou had a picture and you had
like a Van Gogh or a Monet orsomething, and it made that
picture look like a Van Gogh orMonet, right.

Pj Metz (52:40):
Okay. Yeah. Yeah. See, notice? Yeah,

Cassie Breviu (52:44):
that's essentially an AI model that's
trained in a particular style,it can take your image in,
augment it, using somethingcalled the Gann. Again, network,
generative adversarial network,and then create a new image.
Now, though, does the thing thatI built does this an unreal
using Onyx runtime, but itapplies it real time to your

(53:06):
game. So you can make your gameslike in the style of whatever
the model has been trained. Andso Onyx runtime has this open
source library or open modelZoo. So there's tons of
different models there that youcan just go grab and use. And we
have like, like seven differentstyle transfer models, you can
just go graph. And so I grabbedthose, and I started playing

(53:27):
with them and unreal. And this,this took more than two weeks to
get to work, by the way. But itdid work. And it's really cool,
because you can look make itlook like a painting, you can
make it look like a mosaic. Butit's still the vote, the quality
of this experiment that I did isstill not quite good enough to
put in like a production. ButI'm excited to release the code

(53:48):
and like see someone like some,someone that's smarter than me
go figure out how to make ithigher resolution and make it
more realistic. But it issomething that is using Onyx
runtime and inferencing in thegame at every single tick, which
is like kind of insane to thinkabout like that would not be
possible with older hardware at

Brandon Minnick (54:04):
all. Wow. So So if I understand this correctly,
I can almost reskin a game usingthis. So definitely developers
let me add a mod. I can make itlook like I'm anywhere real.
Right? Super

Pj Metz (54:21):
Mario World is now inside of Starry Night.

Cassie Breviu (54:24):
Right, exactly.
Here. I'm going to show you aquick little sneak peek here. Oh
my gosh, yes. With an example.
So I have this really basic,unreal scene. So you can see
it's just like grass and likecement walls, some stone and
some bricks. And then when Iwould hit play on the Unreal
scene, this is applying apicture called rain princess. So

(54:45):
if I go to the Onyx model, zoo,I can show you what the actual
model was trained on. And it'sreally cool to see how that
happens. But this is like whatit was what it would look like
and then when you're able towalk through the world and
mighty This is a very, verysimple game because I just use
like the base one from Unreal.
And then went in and tried tofigure out how to how to make

(55:06):
that work.

Brandon Minnick (55:10):
Incredible.
Just to describe it real quickfor folks listening on the audio
podcast, what we're looking atis Cassius essentially built a
3d world, there's like a brickor concrete fence, there's
almost looks like a, an openingto a well, and some boxes around
this will say a backyard that'sfenced in, and then she's

(55:31):
applied this texture to it. Andnow all those things are still
there, the walls still there,the will still there, the boxes
are still there. But the colors,the design is totally different.
And Cass is pulling up thepicture that it's based on. And
it is on, as I say, unreal,uncanny how similar those

(55:56):
textures became to how thismodel has been trained on.

Cassie Breviu (56:04):
So like this is this is that image, right. And
this is why it's called GreenPrincess model of this image,
that is also like it's what thethis, this painting was called.
So when you look at that, andthen you look back at what this
created.

Pj Metz (56:21):
Yep, impressionist painting of a woman standing in
the rain, and applied that styleto the objects inside of this,
this is amazing.

Brandon Minnick (56:33):
Yet everything like the grass has,

Pj Metz (56:35):
I love seeing this bleeding edge, see.

Cassie Breviu (56:42):
So if I go back to the mosaic example. So this
is the mosaic painting. And if Igo up here, you can see how that
that has translated in theexample of how these models were
built, which are built withpytorch. And then if you go back
to the example you can see, solike this is an example where
like, you can see everything,but the resolution wouldn't be

(57:04):
good enough to like the rain.
For instance, when I feel likeit's like almost good enough
where you could actuallyprobably like, push that. But
the mosaic one, it's like it'sthere, it's really cool. But it
needs a resolution update. SoI've been trying to think about
some different ways that I couldimprove the quality. And I think
I have to go back to the modelitself and think about training
it. One of the things is it'smost computer vision models that
are open source slot, and theygo to 24 by 224 for dimensions.

(57:28):
And this unreal output is alonger triangle or a longer
rectangle. I mean, and so one ofthe things I think is that that
translation might be causing it.
And so if I were to try to traina model with a higher
resolution, I'm wondering if itwould improve that, but But
either way, it's just reallycool to think about like, this

(57:50):
is just that first example ofsomething that you could do and
see where it's gonna go.

Pj Metz (57:55):
Oh, man, I can't wait.

Brandon Minnick (57:58):
It truly looks incredible. The other scenarios
running through my head rightnow just sound amazing where
anybody can just say, hey,here's some artwork, model a
game and all its textures off ofthis insane cast. We only have
about a minute left. Thank youso much for coming on. Before we

(58:20):
leave. let folks know. Where canthey find you folks want to
learn more about Onyx? For folkswho want to follow you on
Twitter? Where can they find youonline?

Cassie Breviu (58:30):
Yeah, so my Twitter handle is @CassieBreviu.
And that heart probably hard tospell but not going to try it,
you'll have to go look it up.
And then the or I'm not going tospell it out. But then the other
way to find me is the Onyxruntime Twitter account, I run
that one as well with some of myteammates. And then also the
YouTube Onyx runtime table orchannel. So it's just YouTube

(58:53):
slash Onyx runtime, oh 1x ONN Xruntime. And that's where I post
lot of my content. And also Ipay attention to what people are
asking for too. So if there'slike something you've been
trying to figure out and youwant a video on it, and you ask
for it, I'm not going to promisebut it will be definitely
something that I'll look at asan indicator of things that
people need. So if you go to theOnyx runtime YouTube channel

(59:14):
now, you'll you'll see videos onjust different ways to actually
build different types ofsolutions, including some onyx
runtime with Xamarin, which Iknow Brandon loves to spend time
for mobile. And specifically, wehave a Xamarin package do that
and C sharp and JavaScript so itkind of shows you how you can do
inferencing and all thesedifferent languages and then the

(59:36):
Unreal Engine One is here, butthe actual demo isn't out yet.
It'll be out tomorrow hopefully

Brandon Minnick (59:46):
amazing. Cassie I'm so glad you were able to
make time for us. Your yourstory is truly truly
inspirational. It's amazing tosee that you can start by just
and centrally being lazy and notwanting to read spreadsheets
anymore to now creating machinelearning models that game

(01:00:07):
developers can use to transformthe industry. Thank you so much
for your time today. And thankyou for joining us for another
episode of eight bits, and we'llsee you next time.