September 9, 2025 • 44 mins
I am always looking for ways to improve. Send me a text and let me know your thoughts! - Kevin
What if your AI could think exactly like you? Not approximating your style based on internet data, but truly capturing your unique patterns of thought, expertise, and decision-making?
Suman Kanaganti, CEO of Personal AI, is creating a future where AI becomes deeply personal—trained on your own memory stack rather than generalized internet knowledge. His company builds small, specialized language models (under 1 billion parameters) that train in minutes yet capture the essence of how individuals or brands think and make decisions.
The difference between these personal language models and general-purpose LLMs like ChatGPT is profound. While large models excel at broad knowledge, they lack the specialized understanding of your unique perspective. Personal AI fills this gap by creating models that see the world through your lens, maintaining complete privacy and data ownership in the process.
This approach is already transforming enterprises, particularly in retail where brands like Wilson are using personal AI for brand-perfect content creation. Financial institutions value the security of on-premises deployment with strict privacy controls. The most fascinating applications emerge when companies create ecosystems of AI personas representing different roles and expertise domains, enabling "compounding collaboration benefits" as these personas interact with each other and human employees.
Looking ahead, Personal AI is developing agent orchestration capabilities and exploring integration with wearable devices—all while maintaining their core principle that your data should belong to you, not platform ecosystems. As Kanaganti puts it: "We want to be at a place where we meet personal AI with the advancements of the data collection methods and then intersect the data to make it ownable by the end user."
Ready to explore how personal AI could transform your business? Visit personalai.com or email Suman directly at s@personalai.com to learn more about this revolutionary approach to artificial intelligence.
🎧 Tech Marketing Rewired is hosted by Kevin Kerner, founder of Mighty & True.
New episodes drop regularly with unfiltered conversations from the frontlines of B2B and tech marketing.
Catch the video version of the podcast on our YouTube channel.
👉 Subscribe, share, and connect with us at www.mightyandtrue.com
📩 Got a guest idea or question? Email us at kevin@mightyandtrue.com
🔗 Follow Kevin on LinkedIn for more insights and behind-the-scenes takes.
📰 Subscribe to Kevin's Substack at: Subscribe to Kevin's Substack at: https://kevinkerner.substack.com/
📰 Want more tech marketing insights? Subscribe to the Drag & Drop newsletter for fresh thinking, real talk, and tools that make your job easier.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Kevin Kerner (00:00):
Hello everyone.
This is Kevin Kerner with TechMarketing Rewired.
I sat down with Suman Kanaganti, ceo of Personal AI, whose
mission is to give people secure, private access to their memory
and knowledge through personallanguage models.
In this episode, we talkedabout how personal AI is
different from largegeneral-purpose LLMs, why
privacy and ownership arecentral to the future of AI, and
(00:22):
how enterprises are alreadyputting personal AI to work for
brand voice, marketing andcompetitive edge.
We dug into his vision foragent orchestration and where
all this goes next as devicesand data collection mature.
So this will be a reallyamazing podcast for those of you
that are really future forwardon small language models.
But before we dive in, techMarketing Rewired is brought to
(00:43):
you by Mighty True languagemodels.
But before we dive in, techMarketing Rewired is brought to
you by Mighty True.
We help B2B marketers movestrategies that get stuck in
execution, making sure that bigideas actually see daylight and
drive results.
Learn more at mightyandtruecom.
All right, let's get to it.
This is Tech Marketing Rewired,suman.
(01:04):
Thank you so much for joiningus here.
Kevin, thanks a lot for havingme.
Yeah, yeah, I'm quite honoredto have you on the show.
In just talking, over the lastfew weeks or so.
I've done quite a bit ofresearch on you, but a lot of
the work that you've done isreally pretty amazing
mission-based work.
I mean your backstory is reallycool.
I wanted to get started.
Can you walk us through yourjourney and how the past work
(01:26):
has sort of led to this missionof personal AI?
Suman Kanuganti (01:31):
My past work, you must be referring to my
previous company called Aira,which was an AI company for
enabling businesses to maketheir locations accessible for
people who are blind and lowvision.
The intention behind thatcompany is around how do you
(01:52):
make the missing visualinformation accessible for
people who cannot see, and AI,of course, played a critical
role, as much as it was 30% AIand 70% human.
During those times, I learned alot about what does it mean to
be having greater access toinformation generally and the
(02:14):
interesting elements of whatinformation that otherwise you
would miss, that helps you makethe decisions that you may not,
you know, foresee.
For example, if Aira was aboutmissing information that is
visual personally is about thatmissing cognitive memory that
otherwise is locked somewheredeep down into your mind.
(02:37):
So what does it mean to begiving access to all that
information and knowledge thatotherwise you have personally
gathered, institutionally andexperienced, and maintain the
relationships, and how do youmake those decisions going
forward, right, so yeah, so alot about it fundamentally is
(02:57):
about access access toinformation, access to yourself,
access to the people around you.
That's kind of the crux ofeverything that we do.
Kevin Kerner (03:07):
Yeah, pretty amazing.
Somewhere I saw a stat, I think, on your site, that 80% of our
memories we forget.
Like there's so much that'sforgotten and it seems like a
mission that's well worthbuilding a solution around
because you can harness thatboth for yourself and also for
others.
It seems like an amazingopportunity, yeah the
interesting.
Suman Kanuganti (03:27):
the interesting thing was, you know, when we
started the company, it was thiscore problem of what would
kevin do, what would suman do,you know, given a specific
context and specific situation?
You know, I had this co-founderof mine at my previous company.
His name is larry bach, I don'tknow in your research he came
up or not.
He passed away.
You know 18 months working withhim.
He was my executive chairmanand he built like 27 different
(03:49):
biotech companies.
He's important to my life,specifically around my
entrepreneurial journey, andonce he passed away because of
pancreatic cancer, we had thismantra of like what would Larry
do?
Right, that was like hugeposters and the idea is, if you
step into those shoes, the wayyou make the decisions is very
(04:11):
specific to you know, theinstitutional knowledge that you
have experienced over a periodof time.
It's not the general knowledge,which is kind of comes back
into the idea of the largelanguage models and the idea of,
you know, personal ai, which isvery much specific to uh and
tailored to individuals yeah,it's early on in the ai.
Kevin Kerner (04:32):
you know, a couple years ago I'd say two or three
years ago, when I started usingit, I heard something about the
use of uh or the um capture ofmemory over time for an
individual, but it was verypersonal.
It was very like you capture itfor your family, for an
individual, but it was verypersonal.
It was very like you capture itfor your family, for your kids.
Imagine, like two or threegenerations down the line, if
you could talk to your greatgrandparent and I'm sure there's
(04:55):
maybe others working on thatproblem.
You've harnessed it for abusiness application purpose,
but it's just fascinating tothink Like I imagine I could
talk to my great-grandfatherover, you know, the next couple
of generations would be prettyamazing.
Suman Kanuganti (05:09):
Absolutely right.
I think the use cases are many,you know, when you start
thinking about personal AIspecifically, you know
individuals comes in all shapesand sizes, right, and different
needs as well.
Our focus always has been thiscore idea of like.
How do we, you know,democratize from the truest
sense?
You know that belongs to you.
(05:30):
Know something that belongs topeople.
It's more like an asset fromour perspective, right.
Can it scale to millions ofpeople or billions of people?
Right?
And how do you architect, youknow, an idea of a model that
thinks like you and make ityours?
Right, an idea of a model thatthinks like you and make it
yours right, there is a greatersense to our vision around the
privacy and the ownership of youknow these models as well.
(05:52):
So the model was important to usbecause otherwise we wouldn't
be fulfilling the promises thatwe set out to do.
And you mentioned, yes, youknow we are focusing on the
business use cases right now.
And you mentioned, yes, we arefocusing on the business use
cases right now, when the uniteconomics are right, when the
market is ready for having apersonal AI for their own, we
(06:13):
foresee a lot of penetration, ifyou will, of having personal
AIs on your devices, on yourphones, on your consumer
robotics that will start, likeleast you know, growing up with
your child as well, right?
So there's lots of futureapplications that fundamentally
can be enabled with personal AIand as much as we are building a
(06:35):
business right now and ourgo-to-market strategy is, you
know this, individual, specific,expert roles.
You know where privacy isimportant, expertise is
important, specialization isimportant.
So, still, like fundamentalcharacteristics of personal AI,
the business model is slightlydifferent.
Kevin Kerner (06:50):
Yeah, yeah, it makes a lot of sense For those
people that might not be theymay be listening or not familiar
with the technical differences.
Can you explain the differencebetween personal LLMs and
general purpose LLMs likeChatGPT, like a lot of people
might think?
Well, I have my history insideof all my chats with ChatGPT,
but what are the broaddifferences between what you're
(07:12):
trying to build and talking to ageneral purpose LLM like
ChatGPT?
Suman Kanuganti (07:18):
What we do is we do not train on external
knowledge as much as wefundamentally train on
individual people's memory stack.
We refer to it as the core, youknow, data and the core
database that goes into thissmall language model.
Architecture, which I'll talkabout in a second, is a
(07:41):
time-based memory that growsover a period of time, and it is
continuous memory, right?
Think about a large languagemodel.
It's pre-trained on acollection of lots of data which
has its specific needs.
They are good at foundationallanguage as well as foundational
general knowledge that isavailable on the internet.
(08:01):
We are talking about creating asmall, discrete,
ultra-specialized,self-contained model that is
more attuned to the likes andthe thinking of an individual
person, right?
So it's kind of the opposite ofa large model.
It's actually a small model.
Even architecturally it is verysmall.
It's less than 1 billionparameters, 1 gigs of size,
(08:22):
trains, less than like in fiveminutes, and it's continuous.
It has its advantages aroundthe unit economics, et cetera,
et cetera and it's designed toscale.
In other words, you wouldexpect the personal AI model to
be running on your device sothat pretty much every person
(08:43):
can have their own personal AImodels.
We are also finding a lot ofapplications around.
You know applying the personalAI technology as well as
platform, in the areas where youknow edge infrastructure is
critical, right.
That is something that iscloser to the people, closer to
the networks, closer to thestadiums right, as compared to
the cloud-based LMs, and thememory that you share with them
(09:06):
is limited to pretty much.
You know that conversation.
Kevin Kerner (09:12):
Are you bringing in though, like if I'm in my
personal LM?
Can I access outsideinformation during that
conversation?
Suman Kanuganti (09:20):
Yeah, it's a good question.
Yes, so, post 2022, what wehave done is we basically
embraced the large languagemodels?
Um, and the way to think aboutit is you are basically seeing
through the lens of you, right?
If you like, let's just say youread an article, right, you
read an article, you absorb thearticle, but you have your own
(09:42):
personal opinion, personal viewof it, right?
So the model is expected tofunction very similarly.
Everything is seen through,like your way of thinking,
because fundamentally, that'sthe filter, which is the PLM,
which is personal language model.
So it's quite fascinating.
I enjoy it quite a bit as well.
Like yesterday, I was doingsome research of my own in
(10:04):
regards to the company and I wastrying to draw some connections
, you know, for personal AI andwith Twilio, right, and that
information is kind of availablein the large language models,
but how, I think, is availablein my personal language model,
right?
So when I have a conversationwith personal language model, it
is, you know, thinking throughthe lens of how I would
(10:26):
otherwise do.
Kevin Kerner (10:27):
Yeah, and then you can interact also with other
people in your organization orother personal AIs, I'm sure?
Yeah, of course.
Suman Kanuganti (10:34):
And then there is like persona to persona, to
interactions as well.
So in the enterprise setting,if you think about it, the whole
concept of collaboration, right, and the agents Agents are very
much you know.
I see companies developingagents for a lot of like
autonomous functions.
If you step back a little bitand think about generally the
(10:57):
workforce that is at the company, right, you have humans that
have specific roles andfunctions and you also have, you
know, some humans that havespecific roles and functions and
you also have some humans thatare experts in a specific lane
within the business, in theproducts or services or strategy
or pricing or competition Imean you name it Several
different functions in thebusiness function.
(11:20):
So personal AI has this conceptof personas.
So think about, like Kevin, aias a unit of persona.
But you can also have a secondpersona that is more tailored
towards your personal life, yourpodcast life, right, so you can
organize the entire workspacewithin a ecosystem of personal
AI into these personas.
(11:40):
So each persona can be arepresentative of an employee's
expertise or a functional role,right.
And now, when you have multipledifferent units of these
personas, where you have humans,you have personas.
You have humans interactingwith the personas, personas
interacting with the personasand personas interacting with
(12:01):
the humans right.
So there's like a collaborationat a compounding, collaboration
at a, you know, compoundingbenefits over a period of time.
Kevin Kerner (12:08):
Yeah, it's an amazing.
That is a mind-blowing concept.
The other thing I was justreading, uh, the five
temptations of a ceo and one ofthe I can't remember which
temptation is, maybe it's numberfour, but it's around um
dissent and trying to trying toum be in an environment where
you get dissent, that you wantdissent to some degree because
you want differing opinions andyou have to create a little bit
(12:28):
of conflict to get thosediffering opinions to make a
good decision yourself, and Iguess you could have
conversations with anotheremployee's expert, ai and really
getting you could get multipleviewpoints coming from multiple
people or you could have theindividual AIs talking to each
other and getting sort of thatbest opinion out for someone to
(12:51):
make a human decision off of.
Suman Kanuganti (12:53):
Right.
You know, in our platform Iguess we'll have a preview of it
towards the end of the podcastwe have a concept of a channel.
It's very similar to like Slackor Microsoft Teams channel.
In the channel you can, ofcourse, have multiple people.
You can also have multiple AIsor AI personas in there, so you
(13:15):
can tag multiple personas andalmost like a panel and get
different perspectives andviewpoints.
So, for example, my CTO hasaround seven core personas.
I have close to 20 corepersonas for myself and
sometimes now you ping differentpeople, you ping Sharon, you
(13:38):
ping Sharon's AI and you pingGates, who is my COO AI, and
kind of see like you know whatare their viewpoints, because
mine is like a lot morestrategic in nature and sharing
is like very specific on youknow what exactly are the
parameters associated with thePLM technology.
So it's quite fascinating.
Kevin Kerner (13:56):
Oh my gosh, that is so cool.
I want access to this.
I mean, I'm the kind of personwho would want access to this.
I would guess there's privacyconcerns, right, but there's
privacy concerns if you put yourdata inside of ChatGPT or
Gemini or whatever.
We work with a lot of companiesthat are in financial and
security and to go through whatwe have, we have to go through
(14:18):
some very strict compliance tobe able to even use AI on their
business.
I would guess there's a bigopportunity in it being very
private, but do you get pushbackalso from people sort of
looking to see how secure youare too?
Are they worried about puttingdata inside this thing?
Suman Kanuganti (14:40):
Well, that's our strength, that's where we
lean in.
Personal AI functions as like aself-contained cluster.
In most cases we also, for someof our highly regulated
customers, like you said, youknow, financial institutions, we
serve banks today.
(15:00):
We deploy, you know, on-prem,you know, on their systems, even
on our cloud.
Every persona has their ownunique memory.
So there is no spillage ofmemory under zero circumstances
to another persona like zero.
So there are privacy layersbuilt at infrastructure layer,
(15:23):
at the application layer, atpersona layer, and then you have
access controls on top of it.
The access controls are quiteinteresting.
When we develop personal AI, ourcore thesis around how AI will
function with people would bethis multiple AI is very similar
to humans, like in acollaborative space,
(15:44):
autonomously are humansupervised.
So if you think about a Slackuser or a human user in a
platform like Slack, we havethese AI personas that access AI
users as well.
So in a way, when we aredeploying into an enterprise
company, we kind of adapt to theexisting controls and policies
(16:07):
access controls and activedirectory policies that they
already have, because eachpersona is going to hierarchical
organization, you know,according to where the team
belongs.
So naturally we kind of, Iguess, extend the existing
policies rather than trying tomake like new policies.
Kevin Kerner (16:24):
Yeah, and when you roll this out across employees,
is there any pushback in termsof them worrying about putting
their persona on a?
I mean, there must be somediscussion that needs to go on
with employees to put theirinformation inside their persona
.
Suman Kanuganti (16:41):
Right.
Kevin Kerner (16:41):
People are pretty willing.
Suman Kanuganti (16:44):
Once they understand where the data goes,
where the data belongs.
Who is the owner of the data?
What's happening behind thescenes?
Of course, we have to betransparent, but that's the
promise for our customers anyway.
So if you look at our privacypolicies or terms of service, or
even customer contracts andagreements, we are pretty clear,
because our business model isnot fundamentally driven by the
(17:05):
data of the customers.
Kevin Kerner (17:08):
Yeah, it's also kind of a no-brainer.
If you want to leverage AI inyour business, this is the best
way to do it, because if you'regoing out and putting it into a
public LLM, well that's theworst case scenario, and you can
certainly try to lock peopleout in their corporate networks,
but who knows how it could beused outside in personal
computers or whatever.
So it's really the best casescenario for an enterprise
(17:31):
that's highly secure to be usingthis stuff.
I know that you just launchedthe Model 3.
I don't think that was launchedrecently.
I'm wondering how I was geekedout of this stuff.
What are the new advances thatyou're excited about that?
Your new Model 3, I don't knowhow new it is, but is now
offering customers.
Suman Kanuganti (17:51):
Yeah, model 3 definitely allowed.
Like multiple personas you knowwork together actively.
What we are seeing is, sinceModel 3, is this compounding
benefit associated with creatingevery new persona that has a
specific role Almost like youknow employees at the company,
(18:12):
because as you have moreemployees, every new persona
that has a specific role, almostlike you know employees at the
company, because as you havemore employees at the company,
then you would start seeing thegrowth rates you know multiply
and increase, right, because youstart having these economies of
scale and then you'recollectively moving forward.
So, with every persona memorycontinuously growing, the value
(18:32):
or the benefit associated withthe ecosystem continues to
exponentially increase, right.
So if you have like one personaversus multiple personas and,
more importantly, all thosepersonas has understanding of
each other, and then now you canget even more creative on
building agents.
That depends on you knowmultiple different functions
(18:55):
across you know departments,right, so that that was an
interesting like observationstarted naturally happening.
You know, in the, in our, inour customer base.
Model 4 is exciting too.
We are in this midst ofprioritization kind of conflicts
(19:20):
right now because we got toprioritize the customers and get
their models and their setupdone.
We are also continuouslypushing our core platform
capabilities from Model 3 toModel 4, and yet keeping up with
the macro developments aroundthe AI.
Anyway, right now, a preview ofModel 4 will be more centered
(19:44):
around creating an agentorchestration layer on top of
these personas, creating like anagent orchestration layer on
top of these personas rightWithin the Persona ecosystem.
Each persona has understandingand intercommunication abilities
, you know.
Between them Now we areadapting to the industry
standards and there is a pusharound, like you know, standards
framework around agent to agentcommunication.
(20:05):
So we want to adapt that forboth our internal persona
communication but also use thesame frameworks for external
agent communication as well.
Yeah Right.
So then you know, you starthaving a standard layer, if you
will, almost like an API fabricthat you would have at the
enterprise companies, you know,back in the days.
Now you have a you know AIfabric or an AI orchestration
(20:26):
fabric on top of it.
So then you have like lots ofinterconnectivity.
That is going on Fascinating,fascinating.
Fascinating Is that with MCP orsome other technology, I cannot
speak to that yet, okay, but wehave a few partners and few
choices and there is someinternal development itself that
we are back contributing tosome of the open source based on
(20:50):
, you know, know, some of thelearnings that we have had.
Kevin Kerner (20:53):
You know, building personal i3 years, yeah we had
two, two things in my mind.
Number one, the applicationsfor this.
It's pretty, it's just prettymind-blowing to talk to you
about it.
And two, just running a, acompany that's in this space,
and how fast everything isevolving.
Just seeing your excitementabout the model 4.
But knowing, knowing that youknow it's got to it.
(21:13):
Both has to keep up, but italso has to be something that's
operationalized and can run andbe secure, which is just
mind-blowing.
I want you to talk through someof the use cases that you see
that have been most inventivefor you just across the business
, and then I want to talk alittle bit about the marketing
use cases, what you seemarketers using this for.
That'll be our primary, ourcore audience for this podcast.
Suman Kanuganti (21:37):
I mean we can start with the marketing use
case.
I believe I can mention thispublicly on the podcast, so we
do are growing in this sector ofretail businesses right now.
So we saw a lot of brandsreferring to the brands, such as
(22:00):
you know, arcteryx, wilson,with you know use cases that has
specific like revenue edge.
Then it kind of moves into usecases that requires like
(22:23):
competitive edge and I'll talkabout it, and then kind of
penetrates into the use casethat has like productivity edge,
right.
So if you think about the usecases, those are the three swim
lanes that we operate in.
For a brand, the revenue edgeis more around how are you able
to make the decisions for aspecific store based on the
(22:44):
customer sentiment that ishappening within the region?
And what are the variables, ordriving variables or you know
driving factors, uh, you know,for for a like a you know a
store productivity analyst or aninventory, uh specialist
persona, right like very deepinto a decision over there kind
of moves the needle on thebalance sheet, right, uh.
(23:04):
So so very high, impacting,high, you know revenue, uh
associated use.
There the competitive edgewithin the retail industry is
more tailored towards.
There is like tough competitionbetween these brands.
You know what is the newproduct that Nike has released
and how does that influence.
You know our products and whatkind of decisions that we need
(23:28):
to make at what period of timeat what regions.
It's a complex operationalchallenge that these companies
has to go through.
So there is a set of personasand use cases to give the
companies competitive edge, andthe last one goes into the
productivity suite.
(23:49):
The productivity is where someof our marketing use cases fall
in as well.
The recent one that I can speakto is a Wilson brand persona.
Right, wilson is very specific,ultra specific, about their
brand and their brandpositioning, their colors, their
ways of speaking about thebrand and let's just say there
(24:10):
is a media interview or theproduct catalog or product
descriptions, every productdescription.
They spend hundreds of dollarsfor one product descriptions,
right, and now they are knockingout the product descriptions
with a Wilson copyright brandpersona which is ultra
specialized, ultra specific toWilson.
Right, that is not influencedby the large language model
(24:30):
trying to borrow some of thecharacteristics from other
Wilson, right, that is notinfluenced by the large language
model.
Trying to borrow some of thecharacteristics from other
brands, right, good point, andthey're extremely happy and
they're knocking down thisproduct descriptions like it's
nobody's business, you know.
So marketing kind of use casesfall into that, like
productivity category and at theproductivity edge.
Kevin Kerner (24:49):
Yeah, wow, super interesting.
And it's all seen through thelens of.
If you're, if you're gatheringthat and you have that internal
data, I would guess that youalso have product or marketing
expertise in the company.
That's the all that informationis being seen through the lens
of.
Like, if I have a really goodproduct marketer who knows
(25:11):
exactly how this stuff should be, and I and I have my own
persona, now I've loaded in allthis data about my brand, the,
the result is going to be a lotbetter than a general llm that,
as you say, is going to pull inall kinds of outside noise,
which it does, gets confused,and you know you also have
compliance.
You know regulatory stuff thatyou that could pull in someone
(25:31):
else's brand, the item, so isthat correct?
So it's both using the internaldata from the company but also
the expertise that's in thecompany around, let's say, brand
or product.
Yeah, if you choose to.
Yeah, it's a choice.
Yeah, what about sales?
And what about the sales sideof things?
Are there sales use cases yet,or do you see sales being a
(25:53):
piece of what gets developed?
Suman Kanuganti (25:59):
Yeah, we do see like sales use cases, but not
the a lot of like AISDRs is veryoperationally driven right,
like what is the sequence ofthings that you would run.
What we see on the sales sideis more centered around you know
how do we.
Let's just say, take a car,like you know, if you have like
two products, right, that arethat has, like you know,
(26:25):
different specifications.
But if a salesperson wererequested to like, hey, I need
to understand the differencesbetween this product versus this
product, or given a certain youknow architectural problem, how
does like my solution kind offits into that solution right,
so it gets into more like adecision-making process that
(26:48):
happens behind the scenes forthe sales, not necessarily like
the outreach idea of, like youknow, like a typical AI SDR.
Yeah, because there's othersolutions that are like targeted
directly at that Right and itgets into like more you know,
integrations and social mediaand taking actions, which kind
of a different tech stack thanwhat we offer.
Kevin Kerner (27:09):
Yeah, it makes sense.
Okay, this is really helpful.
I want to talk a little bitabout your go-to-market strategy
.
At Personalai, I saw you postedabout a lot of congratulations,
a lot of growth, new office, Ithink recently, seven times
growth, larger contracts it allsounds great and I can imagine
it.
Given the product categoryMarketing, this I'll call it
(27:33):
somewhat of a new category ofsmall model AI.
It's got to come with someunique challenges.
How do you approach yourgo-to-market, like what is the
go-to-market you had over thelast year, and how do you see it
changing?
Suman Kanuganti (27:51):
Yeah, I currently have two specific
go-to-market strategies Onewhich I cannot talk about, but
one I will speak openly about,which is our enterprise
go-to-market.
It is true that it is which Icannot talk about, but one I
will speak openly about, whichis our enterprise good market.
It is true that it ischallenging, right?
Specifically, you know, thereis good number of companies
(28:16):
focusing on a lot of specificproblems and going building
vertical solutions and there area lot of companies that also
offer you know services aroundyou know building rack systems
over LLMs and buildinfrastructure solutions inside
the company and primarilynarrows in on a set of highly
(28:50):
regulated sets of use caseswhere the privacy of the data
matters, the precision mattersand the programmability matters.
What is programmability fromour perspective is like?
If you think about these brands, right For them to go through
an AI transformation.
Not every you know brand is like.
Or a retail, for example, likeWalmart, would want to invest in
(29:13):
a lot of their internal AIinfrastructure capabilities,
right, but not a lot of retailswould do, because they are
selling the products.
That's the business, right.
They are designers, so for them, ai transformation optionality
is very, very low.
Selling the products that's thebusiness, right.
They're designers.
So for them, ai transformationoptionality is very, very low,
because they are not ready toinvest in a whole bunch of
(29:35):
services as well as AI MLengineers and develop an AI
group inside right, and that'swhere they need platforms.
They that's where the needplatforms.
The need platforms where theprogrammability is there, which
is I have my business use case.
I know what data influences thedecisions of that business use
(29:57):
case and I can create a personato do exactly that right.
That's the flexibility and theprogrammability without
compromising the you know, theprivacy as well as precision
associated with a specific usecase.
So that's where we are findinga lot of like traction, if you
will, and our go-to-market leansinto creating like brand
(30:21):
awareness around the privacy,precision and programmability.
Kevin Kerner (30:26):
Yeah, you find these companies now have an
issue.
If you hit the right target,they probably have people that
are focused on the tech stackthat's required for their
integration of AI.
They have, like I would imaginenow it's getting mature enough,
there are people at thesecompanies that are actually
(30:46):
looking for solutions like this.
Is that?
Suman Kanuganti (30:48):
right.
Kevin Kerner (30:49):
Yep, yeah, so you can go in with a wedge in the
market to some degree to say,hey, we fit here, right.
This is the exact use case.
Suman Kanuganti (30:58):
Yeah, our wedge right now is growing in the
retail and the entry point isthe COO office.
We do see a lot of strategic IPspecific use cases in there and
from there we kind of scaletoward the company and right now
there are, you know, a fewenterprises that basically is
rolling out to the entireemployees, right, with every
(31:19):
employee having their ownpersona as well.
So those things are happening.
We do some like financialservices.
It's in the early stages and wehave a few banks who are
extremely happy with theprecision of the models as well
as the privacy it offers.
So there's a lot of control andthey are building on personal
AI.
So, I mean, I think, to sum itup right, if you think about,
(31:42):
like large language models, thelarge language models are here
and the use cases are all theway till you know here, right,
and there's a whole bunch ofthings that you would need to
otherwise build in betweenbecause, at the end of the day,
the use cases are very specific.
You know personal, uh, privateip focused, right, you know, for
(32:02):
for those use cases, you eitherget like a you know basically,
development companies or ML AIengineers to fill that gap, or
look at companies who providesthese platforms that are
developer focused, right?
We basically built all that andwe have a platform which is an
application layer and we sellinto the business users, right?
(32:25):
So the go-to-market as well astheir ability to solve their
problem much faster is muchfaster and it's much more
precise, and that's kind of ourkey differentiation out in the
market.
Kevin Kerner (32:39):
Yeah, that's super smart.
Yeah, I was thinking the samething.
You could probably build it ifyou're one of these companies,
but gosh, the time to build it.
If you're into building a AIstack and with things changing
so fast, you might buildsomething.
The next thing you know it'snot.
You know it's useless, right?
Yeah, it'd be really hard to toto compete, so that's, that's
really great.
I want to ask you just a fewquestions.
(32:59):
I have one more.
We'll get into our layer arroulette thing here, but, um,
you mentioned a couple of thingsabout Model 4 and what's going
to change, but I just wanted toask you may not be able to
answer this, but with all thework that you did on the
wearable side you mentioned itat the very beginning I'm
wondering where you see AI in atleast training these AIs with
devices that are surrounded,because I really want that, like
(33:22):
I want that.
So I'm wondering, like, how farout is it that we have some
device that can be listening andlearning for me and training my
personal AI?
Suman Kanuganti (33:32):
Yeah, good question, right, that gets us
into, like the macro.
So one you know track that isdefinitely is the AI.
Of course, like large languagemodel is one piece of it.
Now we have personal languagemodel, kind of filling the gap.
The market is essentiallydemocratizing having access to
(33:54):
the ai.
But then the next question isokay, there is a, there is a
data component to it.
Right, enterprises makes itlike a good entry for companies
like ours because they own thedata, they have specific IP,
right, when you start coming tothe consumers is where you will
start thinking about thisaggregation of data.
(34:16):
The reality that we have livedover the past 20 years, the
internet is the data from apersonal standpoint, from an
individual standpoint, is allsegregated into multiple
different platforms.
It's an unfortunate reality.
So what that means is Kevindoesn't have a memory source
that belongs to him or that hehas his own and, even worse, you
(34:38):
cannot even export the datafrom some of the platforms that
you have.
However, back to the devices,the ubiquity of the devices,
which is your iot devices only,is gonna penetrate more and more
into our lives.
For example, you know, you have, we have our computers, we have
phones in our pockets, we havealexa devices and now we'll have
variables on our body and then,who knows, neural link will be
(34:59):
a real reality in 20 years fromnow, right?
Uh?
When those things happen, it isextremely important for us to
understand where that data goesand who that data belongs to,
right.
For example, you know Apple isbuilding their own ecosystem and
they have their own variables.
Everything is going into Appleecosystem.
As much as they promise you the.
(35:20):
You know the security of it andthe privacy of it.
They still leverage the datafor their own internal business
needs and you know apps-relatedbusiness right.
So what does it mean to theoverall, you know, consumers?
We want to be at a place wherewe meet personal AI with the
(35:42):
advancements of the datacollection methods and then
intersect the data to make itownable by the end user, either
it be a business or anindividual right, because we
want to offer the value with thedata that should belong to them
.
Not necessarily belong in aplatform ecosystem that may have
(36:06):
some other benefits, but that'snot our business model, right?
This is what we want to do withpersonal AI going forward.
So that's like the IoTconnection into personal AI,
which is a future applicationswe will likely evolve into, and
you know we are working with afew partners on that front.
Kevin Kerner (36:22):
It's both the IoT and it's also the data and
privacy and security.
I don't think I'd want to putall of my personal data inside
of an ecosystem that's out therein someone else's ecosystem
that they're training off of,but I would want it in my own,
to just be able to talk to itand learn from it and just query
and have some memory.
(36:44):
So it's really a cool idea.
I hope it happens sooner thanlater.
I'm one of those guys who woulddive into it at the beta.
For sure my wife wouldn't likeit very much.
Suman Kanuganti (36:55):
I mean in a way in 2023,.
We went that direction.
We found the market wasn'tthere.
There was a mismatch in marketexpectations versus the
sophistication of what we weretrying to push, yeah, and there
were a few.
Kevin Kerner (37:09):
There were a few wearables that were going that
route too.
Remember the little clip on theyeah, yeah, and there are pants
.
Suman Kanuganti (37:16):
But what we aim to do is we want to let the IoT
mature a little bit and then wewill have personal AI mature,
you know, to have thoseinference on your devices and
eventually meet at the righttime and the right place.
So, yeah, so we will be playingin that industry eventually.
Kevin Kerner (37:39):
Yes, yeah, I'm a fan, yeah, for sure, ok.
Well, I wanted to.
This has been great.
I could keep going, but I gotto.
I got to let you get back towork and I wanted to do one more
thing with you that we do onthe podcast here.
It's called AI Roulette and I'mgoing to share my screen and
I'm going to go into perplexityor how I have entered this
(38:01):
question.
I'm going to pull this up hereso I can see it here.
This question, I'm going to pullthis up here so I can see it
here.
It says I'm hosting a B2Bmarketing podcast and a little
bit about you, and then give meone bold, unexpected AI roulette
question that I can askmid-interview.
I actually ask it now, shortand punchy, disruptive, don't
give me less than one output,and so I'm going to hit send
(38:23):
here and then I'll have youanswer this and, for the first
time ever, we're going to try tohave your AI answer it.
So let me hit this.
If your AI could perfectlycapture and replay every
conversation you've ever had, doyou think it would make you a
better leader?
Or just expose how often yourepeat yourself?
So let me stop sharing.
Suman Kanuganti (38:39):
Do you mind pasting?
Kevin Kerner (38:40):
that in the chat so that I can copy paste it.
You talked about putting thisinto your own personal ai and
just seeing what it, how itcomes back.
Suman Kanuganti (38:48):
Yeah, what I will do is um, I will uh have my
gates uh answer that.
So let me copy where you aretexting.
I will share my screen and Iwill.
I won't scroll because I Icannot show all the personas
that I have.
Uh, some of them are sensitive,so let's see, click share
button.
Okay, good, on the screen here.
(39:08):
If I scroll up and scroll down,I will have multiple different
personas across the company.
That are close to 60 of themFor this exercise.
I'm gonna go into Gates.
Gates is my I-I-C-O-O and Ijust asked a question about the
team, so that way I don't haveto show my history in here.
But what I will do is I'llsimply paste that question here
(39:31):
and send it to Gates.
Gates is trained on a lot of mycompany strategy-related
activities, so I'm going to lethim speak now.
Kevin Kerner (39:43):
Yeah, yeah, do it.
Suman Kanuganti (39:45):
Leadership enhancement through perfect
memory.
Having perfect conversationalrecall through ai would
undoubtedly make one a betterleader beyond just exposing
repetition patterns.
Here's why pattern recognitionthe ability to analyze
conversation patterns helpsidentify both strengths and
areas for improvement inleadership.
Communication consistency check.
Perfect recall ensuresmessaging remains aligned across
(40:07):
different that's good, refineddecision-making frameworks.
Even if repetition is exposed,it often serves a purpose in
leadership, reinforcing keymessages and ensuring alignment.
(40:29):
The key is using that awarenessto communicate more effectively.
This aligns with our vision atPersonal AI, where we're
building technology thatenhances human capabilities
rather than simply pointing outlimitations.
Our focus on memory,consistency and proven accuracy
makes this possible at scale.
Would you like me to elaborateon?
Kevin Kerner (40:49):
That's really good .
Even if repetition is exposed,it often serves a purpose in
leadership.
Suman Kanuganti (40:57):
So most of the things that otherwise I would
talk about and you can kind ofyou know based on our own
podcast, speaks a lot about howI talk on the perfect memory.
The interesting thing in hereis, if you look at this, you
know personal score 77.
Personal score right.
That is an indication of howspecific or in-depth you know
(41:20):
that response is as it comes tothe way I speak and the way you
know I talk about personal AI.
Kevin Kerner (41:27):
Yeah, that's cool and it really.
The other thing that's reallycool is the voice.
Using the voice of the, you cantrain it on the voice of the
person, which is really makes itmore it.
Just, I don't know it.
Just to me it makes it morepersonal.
Suman Kanuganti (41:40):
Yeah, well, you know, I'm a phone call guy and
when I drive I call my people,when my people are busy, I call
my AI and then have a great time.
It actually helps quite a bit.
I, you know, simply talkthrough the things that are on
top of my mind.
Yeah, it's quite insightful attimes.
(42:00):
Yeah.
Kevin Kerner (42:01):
Yeah, I'm glad someone else does that.
Now I do it with my kids.
When we're in the car we'llpull up ChatGPT and we'll just
ask it all kinds of stuff invoice mode.
Now you do that with personalAI and you will have a history
(42:23):
as well, as it will startshaping the entirety of you.
It just seems like it gives yousuch more perspective than you
could get.
Well, you may not be able toget those conversations that
happen all the time with others,and then you can just dial up
these personas.
Pretty amazing, simon,unbelievable.
I could go on on this stuff.
I can't wait to see what youguys do next.
I'm sure others will want tokeep up with what you're doing.
How should people get a hold ofyou, interact with your company
(42:45):
?
What's the best way to get totalk to you?
Suman Kanuganti (42:48):
That's easy.
Personalai will be the firstone on.
You know, Google search or anyother search.
My email is the letter S atpersonalai, so if you want to
write to me directly, that's thebest way.
Kevin Kerner (43:02):
Awesome.
Can you interact?
Do you expose any of the AIs toexternally?
Do you have any AIs out therethat people can talk to?
Suman Kanuganti (43:10):
There are some on personal AI Okay, and most of
our work is very much ultraprivate.
So, it's less external usecases but a lot internal use
cases, but we try to expose someof them just for people to play
around yeah, you're not tryingto build character dot AI or
(43:32):
anything silly no, no, in fact,you know, in 2023 there was
almost like a competitivelandscape between inflection AI,
character AI and personal AI.
But our approach has been kindof starts, you know, starts from
the people, necessarily.
You know, simply talk to theprominent figures, right.
(43:53):
Our focus was more centeredaround, like, the individual who
is actually getting on personalAI.
So there was confusion.
It's like, oh, this ischaracter AI.
No, it's not.
It's not character AI.
Kevin Kerner (44:03):
Yeah, yeah, and you're on the right path.
I wouldn't.
Yeah, it's characterized awhole different, whole different
ball of wax.
But yeah, it's, Suman, I'm sohappy that you joined that we,
that I reached out and youjoined us.
I really can't thank you enough.
This has been awesome and I hadfun.
Suman Kanuganti (44:22):
Kevin thanks a lot for doing this.
Thank you so much.
Hello everyone.
This is Kevin Kerner with TechMarketing Rewired.
I sat down with Suman Kanaganti, ceo of Personal AI, whose
mission is to give people secure, private access to their memory
and knowledge through personallanguage models.
In this episode, we talkedabout how personal AI is
different from largegeneral-purpose LLMs, why
privacy and ownership arecentral to the future of AI, and
(00:22):
how enterprises are alreadyputting personal AI to work for
brand voice, marketing andcompetitive edge.
We dug into his vision foragent orchestration and where
all this goes next as devicesand data collection mature.
So this will be a reallyamazing podcast for those of you
that are really future forwardon small language models.
But before we dive in, techMarketing Rewired is brought to
(00:43):
you by Mighty True languagemodels.
But before we dive in, techMarketing Rewired is brought to
you by Mighty True.
We help B2B marketers movestrategies that get stuck in
execution, making sure that bigideas actually see daylight and
drive results.
Learn more at mightyandtruecom.
All right, let's get to it.
This is Tech Marketing Rewired,suman.
(01:04):
Thank you so much for joiningus here.
Kevin, thanks a lot for havingme.
Yeah, yeah, I'm quite honoredto have you on the show.
In just talking, over the lastfew weeks or so.
I've done quite a bit ofresearch on you, but a lot of
the work that you've done isreally pretty amazing
mission-based work.
I mean your backstory is reallycool.
I wanted to get started.
Can you walk us through yourjourney and how the past work
(01:26):
has sort of led to this missionof personal AI?
Suman Kanuganti (01:31):
My past work, you must be referring to my
previous company called Aira,which was an AI company for
enabling businesses to maketheir locations accessible for
people who are blind and lowvision.
The intention behind thatcompany is around how do you
(01:52):
make the missing visualinformation accessible for
people who cannot see, and AI,of course, played a critical
role, as much as it was 30% AIand 70% human.
During those times, I learned alot about what does it mean to
be having greater access toinformation generally and the
(02:14):
interesting elements of whatinformation that otherwise you
would miss, that helps you makethe decisions that you may not,
you know, foresee.
For example, if Aira was aboutmissing information that is
visual personally is about thatmissing cognitive memory that
otherwise is locked somewheredeep down into your mind.
(02:37):
So what does it mean to begiving access to all that
information and knowledge thatotherwise you have personally
gathered, institutionally andexperienced, and maintain the
relationships, and how do youmake those decisions going
forward, right, so yeah, so alot about it fundamentally is
(02:57):
about access access toinformation, access to yourself,
access to the people around you.
That's kind of the crux ofeverything that we do.
Kevin Kerner (03:07):
Yeah, pretty amazing.
Somewhere I saw a stat, I think, on your site, that 80% of our
memories we forget.
Like there's so much that'sforgotten and it seems like a
mission that's well worthbuilding a solution around
because you can harness thatboth for yourself and also for
others.
It seems like an amazingopportunity, yeah the
interesting.
Suman Kanuganti (03:27):
the interesting thing was, you know, when we
started the company, it was thiscore problem of what would
kevin do, what would suman do,you know, given a specific
context and specific situation?
You know, I had this co-founderof mine at my previous company.
His name is larry bach, I don'tknow in your research he came
up or not.
He passed away.
You know 18 months working withhim.
He was my executive chairmanand he built like 27 different
(03:49):
biotech companies.
He's important to my life,specifically around my
entrepreneurial journey, andonce he passed away because of
pancreatic cancer, we had thismantra of like what would Larry
do?
Right, that was like hugeposters and the idea is, if you
step into those shoes, the wayyou make the decisions is very
(04:11):
specific to you know, theinstitutional knowledge that you
have experienced over a periodof time.
It's not the general knowledge,which is kind of comes back
into the idea of the largelanguage models and the idea of,
you know, personal ai, which isvery much specific to uh and
tailored to individuals yeah,it's early on in the ai.
Kevin Kerner (04:32):
you know, a couple years ago I'd say two or three
years ago, when I started usingit, I heard something about the
use of uh or the um capture ofmemory over time for an
individual, but it was verypersonal.
It was very like you capture itfor your family, for an
individual, but it was verypersonal.
It was very like you capture itfor your family, for your kids.
Imagine, like two or threegenerations down the line, if
you could talk to your greatgrandparent and I'm sure there's
(04:55):
maybe others working on thatproblem.
You've harnessed it for abusiness application purpose,
but it's just fascinating tothink Like I imagine I could
talk to my great-grandfatherover, you know, the next couple
of generations would be prettyamazing.
Suman Kanuganti (05:09):
Absolutely right.
I think the use cases are many,you know, when you start
thinking about personal AIspecifically, you know
individuals comes in all shapesand sizes, right, and different
needs as well.
Our focus always has been thiscore idea of like.
How do we, you know,democratize from the truest
sense?
You know that belongs to you.
(05:30):
Know something that belongs topeople.
It's more like an asset fromour perspective, right.
Can it scale to millions ofpeople or billions of people?
Right?
And how do you architect, youknow, an idea of a model that
thinks like you and make ityours?
Right, an idea of a model thatthinks like you and make it
yours right, there is a greatersense to our vision around the
privacy and the ownership of youknow these models as well.
(05:52):
So the model was important to usbecause otherwise we wouldn't
be fulfilling the promises thatwe set out to do.
And you mentioned, yes, youknow we are focusing on the
business use cases right now.
And you mentioned, yes, we arefocusing on the business use
cases right now, when the uniteconomics are right, when the
market is ready for having apersonal AI for their own, we
(06:13):
foresee a lot of penetration, ifyou will, of having personal
AIs on your devices, on yourphones, on your consumer
robotics that will start, likeleast you know, growing up with
your child as well, right?
So there's lots of futureapplications that fundamentally
can be enabled with personal AIand as much as we are building a
(06:35):
business right now and ourgo-to-market strategy is, you
know this, individual, specific,expert roles.
You know where privacy isimportant, expertise is
important, specialization isimportant.
So, still, like fundamentalcharacteristics of personal AI,
the business model is slightlydifferent.
Kevin Kerner (06:50):
Yeah, yeah, it makes a lot of sense For those
people that might not be theymay be listening or not familiar
with the technical differences.
Can you explain the differencebetween personal LLMs and
general purpose LLMs likeChatGPT, like a lot of people
might think?
Well, I have my history insideof all my chats with ChatGPT,
but what are the broaddifferences between what you're
(07:12):
trying to build and talking to ageneral purpose LLM like
ChatGPT?
Suman Kanuganti (07:18):
What we do is we do not train on external
knowledge as much as wefundamentally train on
individual people's memory stack.
We refer to it as the core, youknow, data and the core
database that goes into thissmall language model.
Architecture, which I'll talkabout in a second, is a
(07:41):
time-based memory that growsover a period of time, and it is
continuous memory, right?
Think about a large languagemodel.
It's pre-trained on acollection of lots of data which
has its specific needs.
They are good at foundationallanguage as well as foundational
general knowledge that isavailable on the internet.
(08:01):
We are talking about creating asmall, discrete,
ultra-specialized,self-contained model that is
more attuned to the likes andthe thinking of an individual
person, right?
So it's kind of the opposite ofa large model.
It's actually a small model.
Even architecturally it is verysmall.
It's less than 1 billionparameters, 1 gigs of size,
(08:22):
trains, less than like in fiveminutes, and it's continuous.
It has its advantages aroundthe unit economics, et cetera,
et cetera and it's designed toscale.
In other words, you wouldexpect the personal AI model to
be running on your device sothat pretty much every person
(08:43):
can have their own personal AImodels.
We are also finding a lot ofapplications around.
You know applying the personalAI technology as well as
platform, in the areas where youknow edge infrastructure is
critical, right.
That is something that iscloser to the people, closer to
the networks, closer to thestadiums right, as compared to
the cloud-based LMs, and thememory that you share with them
(09:06):
is limited to pretty much.
You know that conversation.
Kevin Kerner (09:12):
Are you bringing in though, like if I'm in my
personal LM?
Can I access outsideinformation during that
conversation?
Suman Kanuganti (09:20):
Yeah, it's a good question.
Yes, so, post 2022, what wehave done is we basically
embraced the large languagemodels?
Um, and the way to think aboutit is you are basically seeing
through the lens of you, right?
If you like, let's just say youread an article, right, you
read an article, you absorb thearticle, but you have your own
(09:42):
personal opinion, personal viewof it, right?
So the model is expected tofunction very similarly.
Everything is seen through,like your way of thinking,
because fundamentally, that'sthe filter, which is the PLM,
which is personal language model.
So it's quite fascinating.
I enjoy it quite a bit as well.
Like yesterday, I was doingsome research of my own in
(10:04):
regards to the company and I wastrying to draw some connections
, you know, for personal AI andwith Twilio, right, and that
information is kind of availablein the large language models,
but how, I think, is availablein my personal language model,
right?
So when I have a conversationwith personal language model, it
is, you know, thinking throughthe lens of how I would
(10:26):
otherwise do.
Kevin Kerner (10:27):
Yeah, and then you can interact also with other
people in your organization orother personal AIs, I'm sure?
Yeah, of course.
Suman Kanuganti (10:34):
And then there is like persona to persona, to
interactions as well.
So in the enterprise setting,if you think about it, the whole
concept of collaboration, right, and the agents Agents are very
much you know.
I see companies developingagents for a lot of like
autonomous functions.
If you step back a little bitand think about generally the
(10:57):
workforce that is at the company, right, you have humans that
have specific roles andfunctions and you also have, you
know, some humans that havespecific roles and functions and
you also have some humans thatare experts in a specific lane
within the business, in theproducts or services or strategy
or pricing or competition Imean you name it Several
different functions in thebusiness function.
(11:20):
So personal AI has this conceptof personas.
So think about, like Kevin, aias a unit of persona.
But you can also have a secondpersona that is more tailored
towards your personal life, yourpodcast life, right, so you can
organize the entire workspacewithin a ecosystem of personal
AI into these personas.
(11:40):
So each persona can be arepresentative of an employee's
expertise or a functional role,right.
And now, when you have multipledifferent units of these
personas, where you have humans,you have personas.
You have humans interactingwith the personas, personas
interacting with the personasand personas interacting with
(12:01):
the humans right.
So there's like a collaborationat a compounding, collaboration
at a, you know, compoundingbenefits over a period of time.
Kevin Kerner (12:08):
Yeah, it's an amazing.
That is a mind-blowing concept.
The other thing I was justreading, uh, the five
temptations of a ceo and one ofthe I can't remember which
temptation is, maybe it's numberfour, but it's around um
dissent and trying to trying toum be in an environment where
you get dissent, that you wantdissent to some degree because
you want differing opinions andyou have to create a little bit
(12:28):
of conflict to get thosediffering opinions to make a
good decision yourself, and Iguess you could have
conversations with anotheremployee's expert, ai and really
getting you could get multipleviewpoints coming from multiple
people or you could have theindividual AIs talking to each
other and getting sort of thatbest opinion out for someone to
(12:51):
make a human decision off of.
Suman Kanuganti (12:53):
Right.
You know, in our platform Iguess we'll have a preview of it
towards the end of the podcastwe have a concept of a channel.
It's very similar to like Slackor Microsoft Teams channel.
In the channel you can, ofcourse, have multiple people.
You can also have multiple AIsor AI personas in there, so you
(13:15):
can tag multiple personas andalmost like a panel and get
different perspectives andviewpoints.
So, for example, my CTO hasaround seven core personas.
I have close to 20 corepersonas for myself and
sometimes now you ping differentpeople, you ping Sharon, you
(13:38):
ping Sharon's AI and you pingGates, who is my COO AI, and
kind of see like you know whatare their viewpoints, because
mine is like a lot morestrategic in nature and sharing
is like very specific on youknow what exactly are the
parameters associated with thePLM technology.
So it's quite fascinating.
Kevin Kerner (13:56):
Oh my gosh, that is so cool.
I want access to this.
I mean, I'm the kind of personwho would want access to this.
I would guess there's privacyconcerns, right, but there's
privacy concerns if you put yourdata inside of ChatGPT or
Gemini or whatever.
We work with a lot of companiesthat are in financial and
security and to go through whatwe have, we have to go through
(14:18):
some very strict compliance tobe able to even use AI on their
business.
I would guess there's a bigopportunity in it being very
private, but do you get pushbackalso from people sort of
looking to see how secure youare too?
Are they worried about puttingdata inside this thing?
Suman Kanuganti (14:40):
Well, that's our strength, that's where we
lean in.
Personal AI functions as like aself-contained cluster.
In most cases we also, for someof our highly regulated
customers, like you said, youknow, financial institutions, we
serve banks today.
(15:00):
We deploy, you know, on-prem,you know, on their systems, even
on our cloud.
Every persona has their ownunique memory.
So there is no spillage ofmemory under zero circumstances
to another persona like zero.
So there are privacy layersbuilt at infrastructure layer,
(15:23):
at the application layer, atpersona layer, and then you have
access controls on top of it.
The access controls are quiteinteresting.
When we develop personal AI, ourcore thesis around how AI will
function with people would bethis multiple AI is very similar
to humans, like in acollaborative space,
(15:44):
autonomously are humansupervised.
So if you think about a Slackuser or a human user in a
platform like Slack, we havethese AI personas that access AI
users as well.
So in a way, when we aredeploying into an enterprise
company, we kind of adapt to theexisting controls and policies
(16:07):
access controls and activedirectory policies that they
already have, because eachpersona is going to hierarchical
organization, you know,according to where the team
belongs.
So naturally we kind of, Iguess, extend the existing
policies rather than trying tomake like new policies.
Kevin Kerner (16:24):
Yeah, and when you roll this out across employees,
is there any pushback in termsof them worrying about putting
their persona on a?
I mean, there must be somediscussion that needs to go on
with employees to put theirinformation inside their persona
.
Suman Kanuganti (16:41):
Right.
Kevin Kerner (16:41):
People are pretty willing.
Suman Kanuganti (16:44):
Once they understand where the data goes,
where the data belongs.
Who is the owner of the data?
What's happening behind thescenes?
Of course, we have to betransparent, but that's the
promise for our customers anyway.
So if you look at our privacypolicies or terms of service, or
even customer contracts andagreements, we are pretty clear,
because our business model isnot fundamentally driven by the
(17:05):
data of the customers.
Kevin Kerner (17:08):
Yeah, it's also kind of a no-brainer.
If you want to leverage AI inyour business, this is the best
way to do it, because if you'regoing out and putting it into a
public LLM, well that's theworst case scenario, and you can
certainly try to lock peopleout in their corporate networks,
but who knows how it could beused outside in personal
computers or whatever.
So it's really the best casescenario for an enterprise
(17:31):
that's highly secure to be usingthis stuff.
I know that you just launchedthe Model 3.
I don't think that was launchedrecently.
I'm wondering how I was geekedout of this stuff.
What are the new advances thatyou're excited about that?
Your new Model 3, I don't knowhow new it is, but is now
offering customers.
Suman Kanuganti (17:51):
Yeah, model 3 definitely allowed.
Like multiple personas you knowwork together actively.
What we are seeing is, sinceModel 3, is this compounding
benefit associated with creatingevery new persona that has a
specific role Almost like youknow employees at the company,
(18:12):
because as you have moreemployees, every new persona
that has a specific role, almostlike you know employees at the
company, because as you havemore employees at the company,
then you would start seeing thegrowth rates you know multiply
and increase, right, because youstart having these economies of
scale and then you'recollectively moving forward.
So, with every persona memorycontinuously growing, the value
(18:32):
or the benefit associated withthe ecosystem continues to
exponentially increase, right.
So if you have like one personaversus multiple personas and,
more importantly, all thosepersonas has understanding of
each other, and then now you canget even more creative on
building agents.
That depends on you knowmultiple different functions
(18:55):
across you know departments,right, so that that was an
interesting like observationstarted naturally happening.
You know, in the, in our, inour customer base.
Model 4 is exciting too.
We are in this midst ofprioritization kind of conflicts
(19:20):
right now because we got toprioritize the customers and get
their models and their setupdone.
We are also continuouslypushing our core platform
capabilities from Model 3 toModel 4, and yet keeping up with
the macro developments aroundthe AI.
Anyway, right now, a preview ofModel 4 will be more centered
(19:44):
around creating an agentorchestration layer on top of
these personas, creating like anagent orchestration layer on
top of these personas rightWithin the Persona ecosystem.
Each persona has understandingand intercommunication abilities
, you know.
Between them Now we areadapting to the industry
standards and there is a pusharound, like you know, standards
framework around agent to agentcommunication.
(20:05):
So we want to adapt that forboth our internal persona
communication but also use thesame frameworks for external
agent communication as well.
Yeah Right.
So then you know, you starthaving a standard layer, if you
will, almost like an API fabricthat you would have at the
enterprise companies, you know,back in the days.
Now you have a you know AIfabric or an AI orchestration
(20:26):
fabric on top of it.
So then you have like lots ofinterconnectivity.
That is going on Fascinating,fascinating.
Fascinating Is that with MCP orsome other technology, I cannot
speak to that yet, okay, but wehave a few partners and few
choices and there is someinternal development itself that
we are back contributing tosome of the open source based on
(20:50):
, you know, know, some of thelearnings that we have had.
Kevin Kerner (20:53):
You know, building personal i3 years, yeah we had
two, two things in my mind.
Number one, the applicationsfor this.
It's pretty, it's just prettymind-blowing to talk to you
about it.
And two, just running a, acompany that's in this space,
and how fast everything isevolving.
Just seeing your excitementabout the model 4.
But knowing, knowing that youknow it's got to it.
(21:13):
Both has to keep up, but italso has to be something that's
operationalized and can run andbe secure, which is just
mind-blowing.
I want you to talk through someof the use cases that you see
that have been most inventivefor you just across the business
, and then I want to talk alittle bit about the marketing
use cases, what you seemarketers using this for.
That'll be our primary, ourcore audience for this podcast.
Suman Kanuganti (21:37):
I mean we can start with the marketing use
case.
I believe I can mention thispublicly on the podcast, so we
do are growing in this sector ofretail businesses right now.
So we saw a lot of brandsreferring to the brands, such as
(22:00):
you know, arcteryx, wilson,with you know use cases that has
specific like revenue edge.
Then it kind of moves into usecases that requires like
(22:23):
competitive edge and I'll talkabout it, and then kind of
penetrates into the use casethat has like productivity edge,
right.
So if you think about the usecases, those are the three swim
lanes that we operate in.
For a brand, the revenue edgeis more around how are you able
to make the decisions for aspecific store based on the
(22:44):
customer sentiment that ishappening within the region?
And what are the variables, ordriving variables or you know
driving factors, uh, you know,for for a like a you know a
store productivity analyst or aninventory, uh specialist
persona, right like very deepinto a decision over there kind
of moves the needle on thebalance sheet, right, uh.
(23:04):
So so very high, impacting,high, you know revenue, uh
associated use.
There the competitive edgewithin the retail industry is
more tailored towards.
There is like tough competitionbetween these brands.
You know what is the newproduct that Nike has released
and how does that influence.
You know our products and whatkind of decisions that we need
(23:28):
to make at what period of timeat what regions.
It's a complex operationalchallenge that these companies
has to go through.
So there is a set of personasand use cases to give the
companies competitive edge, andthe last one goes into the
productivity suite.
(23:49):
The productivity is where someof our marketing use cases fall
in as well.
The recent one that I can speakto is a Wilson brand persona.
Right, wilson is very specific,ultra specific, about their
brand and their brandpositioning, their colors, their
ways of speaking about thebrand and let's just say there
(24:10):
is a media interview or theproduct catalog or product
descriptions, every productdescription.
They spend hundreds of dollarsfor one product descriptions,
right, and now they are knockingout the product descriptions
with a Wilson copyright brandpersona which is ultra
specialized, ultra specific toWilson.
Right, that is not influencedby the large language model
(24:30):
trying to borrow some of thecharacteristics from other
Wilson, right, that is notinfluenced by the large language
model.
Trying to borrow some of thecharacteristics from other
brands, right, good point, andthey're extremely happy and
they're knocking down thisproduct descriptions like it's
nobody's business, you know.
So marketing kind of use casesfall into that, like
productivity category and at theproductivity edge.
Kevin Kerner (24:49):
Yeah, wow, super interesting.
And it's all seen through thelens of.
If you're, if you're gatheringthat and you have that internal
data, I would guess that youalso have product or marketing
expertise in the company.
That's the all that informationis being seen through the lens
of.
Like, if I have a really goodproduct marketer who knows
(25:11):
exactly how this stuff should be, and I and I have my own
persona, now I've loaded in allthis data about my brand, the,
the result is going to be a lotbetter than a general llm that,
as you say, is going to pull inall kinds of outside noise,
which it does, gets confused,and you know you also have
compliance.
You know regulatory stuff thatyou that could pull in someone
(25:31):
else's brand, the item, so isthat correct?
So it's both using the internaldata from the company but also
the expertise that's in thecompany around, let's say, brand
or product.
Yeah, if you choose to.
Yeah, it's a choice.
Yeah, what about sales?
And what about the sales sideof things?
Are there sales use cases yet,or do you see sales being a
(25:53):
piece of what gets developed?
Suman Kanuganti (25:59):
Yeah, we do see like sales use cases, but not
the a lot of like AISDRs is veryoperationally driven right,
like what is the sequence ofthings that you would run.
What we see on the sales sideis more centered around you know
how do we.
Let's just say, take a car,like you know, if you have like
two products, right, that arethat has, like you know,
(26:25):
different specifications.
But if a salesperson wererequested to like, hey, I need
to understand the differencesbetween this product versus this
product, or given a certain youknow architectural problem, how
does like my solution kind offits into that solution right,
so it gets into more like adecision-making process that
(26:48):
happens behind the scenes forthe sales, not necessarily like
the outreach idea of, like youknow, like a typical AI SDR.
Yeah, because there's othersolutions that are like targeted
directly at that Right and itgets into like more you know,
integrations and social mediaand taking actions, which kind
of a different tech stack thanwhat we offer.
Kevin Kerner (27:09):
Yeah, it makes sense.
Okay, this is really helpful.
I want to talk a little bitabout your go-to-market strategy
.
At Personalai, I saw you postedabout a lot of congratulations,
a lot of growth, new office, Ithink recently, seven times
growth, larger contracts it allsounds great and I can imagine
it.
Given the product categoryMarketing, this I'll call it
(27:33):
somewhat of a new category ofsmall model AI.
It's got to come with someunique challenges.
How do you approach yourgo-to-market, like what is the
go-to-market you had over thelast year, and how do you see it
changing?
Suman Kanuganti (27:51):
Yeah, I currently have two specific
go-to-market strategies Onewhich I cannot talk about, but
one I will speak openly about,which is our enterprise
go-to-market.
It is true that it is which Icannot talk about, but one I
will speak openly about, whichis our enterprise good market.
It is true that it ischallenging, right?
Specifically, you know, thereis good number of companies
(28:16):
focusing on a lot of specificproblems and going building
vertical solutions and there area lot of companies that also
offer you know services aroundyou know building rack systems
over LLMs and buildinfrastructure solutions inside
the company and primarilynarrows in on a set of highly
(28:50):
regulated sets of use caseswhere the privacy of the data
matters, the precision mattersand the programmability matters.
What is programmability fromour perspective is like?
If you think about these brands, right For them to go through
an AI transformation.
Not every you know brand is like.
Or a retail, for example, likeWalmart, would want to invest in
(29:13):
a lot of their internal AIinfrastructure capabilities,
right, but not a lot of retailswould do, because they are
selling the products.
That's the business, right.
They are designers, so for them, ai transformation optionality
is very, very low.
Selling the products that's thebusiness, right.
They're designers.
So for them, ai transformationoptionality is very, very low,
because they are not ready toinvest in a whole bunch of
(29:35):
services as well as AI MLengineers and develop an AI
group inside right, and that'swhere they need platforms.
They that's where the needplatforms.
The need platforms where theprogrammability is there, which
is I have my business use case.
I know what data influences thedecisions of that business use
(29:57):
case and I can create a personato do exactly that right.
That's the flexibility and theprogrammability without
compromising the you know, theprivacy as well as precision
associated with a specific usecase.
So that's where we are findinga lot of like traction, if you
will, and our go-to-market leansinto creating like brand
(30:21):
awareness around the privacy,precision and programmability.
Kevin Kerner (30:26):
Yeah, you find these companies now have an
issue.
If you hit the right target,they probably have people that
are focused on the tech stackthat's required for their
integration of AI.
They have, like I would imaginenow it's getting mature enough,
there are people at thesecompanies that are actually
(30:46):
looking for solutions like this.
Is that?
Suman Kanuganti (30:48):
right.
Kevin Kerner (30:49):
Yep, yeah, so you can go in with a wedge in the
market to some degree to say,hey, we fit here, right.
This is the exact use case.
Suman Kanuganti (30:58):
Yeah, our wedge right now is growing in the
retail and the entry point isthe COO office.
We do see a lot of strategic IPspecific use cases in there and
from there we kind of scaletoward the company and right now
there are, you know, a fewenterprises that basically is
rolling out to the entireemployees, right, with every
(31:19):
employee having their ownpersona as well.
So those things are happening.
We do some like financialservices.
It's in the early stages and wehave a few banks who are
extremely happy with theprecision of the models as well
as the privacy it offers.
So there's a lot of control andthey are building on personal
AI.
So, I mean, I think, to sum itup right, if you think about,
(31:42):
like large language models, thelarge language models are here
and the use cases are all theway till you know here, right,
and there's a whole bunch ofthings that you would need to
otherwise build in betweenbecause, at the end of the day,
the use cases are very specific.
You know personal, uh, privateip focused, right, you know, for
(32:02):
for those use cases, you eitherget like a you know basically,
development companies or ML AIengineers to fill that gap, or
look at companies who providesthese platforms that are
developer focused, right?
We basically built all that andwe have a platform which is an
application layer and we sellinto the business users, right?
(32:25):
So the go-to-market as well astheir ability to solve their
problem much faster is muchfaster and it's much more
precise, and that's kind of ourkey differentiation out in the
market.
Kevin Kerner (32:39):
Yeah, that's super smart.
Yeah, I was thinking the samething.
You could probably build it ifyou're one of these companies,
but gosh, the time to build it.
If you're into building a AIstack and with things changing
so fast, you might buildsomething.
The next thing you know it'snot.
You know it's useless, right?
Yeah, it'd be really hard to toto compete, so that's, that's
really great.
I want to ask you just a fewquestions.
(32:59):
I have one more.
We'll get into our layer arroulette thing here, but, um,
you mentioned a couple of thingsabout Model 4 and what's going
to change, but I just wanted toask you may not be able to
answer this, but with all thework that you did on the
wearable side you mentioned itat the very beginning I'm
wondering where you see AI in atleast training these AIs with
devices that are surrounded,because I really want that, like
(33:22):
I want that.
So I'm wondering, like, how farout is it that we have some
device that can be listening andlearning for me and training my
personal AI?
Suman Kanuganti (33:32):
Yeah, good question, right, that gets us
into, like the macro.
So one you know track that isdefinitely is the AI.
Of course, like large languagemodel is one piece of it.
Now we have personal languagemodel, kind of filling the gap.
The market is essentiallydemocratizing having access to
(33:54):
the ai.
But then the next question isokay, there is a, there is a
data component to it.
Right, enterprises makes itlike a good entry for companies
like ours because they own thedata, they have specific IP,
right, when you start coming tothe consumers is where you will
start thinking about thisaggregation of data.
(34:16):
The reality that we have livedover the past 20 years, the
internet is the data from apersonal standpoint, from an
individual standpoint, is allsegregated into multiple
different platforms.
It's an unfortunate reality.
So what that means is Kevindoesn't have a memory source
that belongs to him or that hehas his own and, even worse, you
(34:38):
cannot even export the datafrom some of the platforms that
you have.
However, back to the devices,the ubiquity of the devices,
which is your iot devices only,is gonna penetrate more and more
into our lives.
For example, you know, you have, we have our computers, we have
phones in our pockets, we havealexa devices and now we'll have
variables on our body and then,who knows, neural link will be
(34:59):
a real reality in 20 years fromnow, right?
Uh?
When those things happen, it isextremely important for us to
understand where that data goesand who that data belongs to,
right.
For example, you know Apple isbuilding their own ecosystem and
they have their own variables.
Everything is going into Appleecosystem.
As much as they promise you the.
(35:20):
You know the security of it andthe privacy of it.
They still leverage the datafor their own internal business
needs and you know apps-relatedbusiness right.
So what does it mean to theoverall, you know, consumers?
We want to be at a place wherewe meet personal AI with the
(35:42):
advancements of the datacollection methods and then
intersect the data to make itownable by the end user, either
it be a business or anindividual right, because we
want to offer the value with thedata that should belong to them
.
Not necessarily belong in aplatform ecosystem that may have
(36:06):
some other benefits, but that'snot our business model, right?
This is what we want to do withpersonal AI going forward.
So that's like the IoTconnection into personal AI,
which is a future applicationswe will likely evolve into, and
you know we are working with afew partners on that front.
Kevin Kerner (36:22):
It's both the IoT and it's also the data and
privacy and security.
I don't think I'd want to putall of my personal data inside
of an ecosystem that's out therein someone else's ecosystem
that they're training off of,but I would want it in my own,
to just be able to talk to itand learn from it and just query
and have some memory.
(36:44):
So it's really a cool idea.
I hope it happens sooner thanlater.
I'm one of those guys who woulddive into it at the beta.
For sure my wife wouldn't likeit very much.
Suman Kanuganti (36:55):
I mean in a way in 2023,.
We went that direction.
We found the market wasn'tthere.
There was a mismatch in marketexpectations versus the
sophistication of what we weretrying to push, yeah, and there
were a few.
Kevin Kerner (37:09):
There were a few wearables that were going that
route too.
Remember the little clip on theyeah, yeah, and there are pants
.
Suman Kanuganti (37:16):
But what we aim to do is we want to let the IoT
mature a little bit and then wewill have personal AI mature,
you know, to have thoseinference on your devices and
eventually meet at the righttime and the right place.
So, yeah, so we will be playingin that industry eventually.
Kevin Kerner (37:39):
Yes, yeah, I'm a fan, yeah, for sure, ok.
Well, I wanted to.
This has been great.
I could keep going, but I gotto.
I got to let you get back towork and I wanted to do one more
thing with you that we do onthe podcast here.
It's called AI Roulette and I'mgoing to share my screen and
I'm going to go into perplexityor how I have entered this
(38:01):
question.
I'm going to pull this up hereso I can see it here.
This question, I'm going to pullthis up here so I can see it
here.
It says I'm hosting a B2Bmarketing podcast and a little
bit about you, and then give meone bold, unexpected AI roulette
question that I can askmid-interview.
I actually ask it now, shortand punchy, disruptive, don't
give me less than one output,and so I'm going to hit send
(38:23):
here and then I'll have youanswer this and, for the first
time ever, we're going to try tohave your AI answer it.
So let me hit this.
If your AI could perfectlycapture and replay every
conversation you've ever had, doyou think it would make you a
better leader?
Or just expose how often yourepeat yourself?
So let me stop sharing.
Suman Kanuganti (38:39):
Do you mind pasting?
Kevin Kerner (38:40):
that in the chat so that I can copy paste it.
You talked about putting thisinto your own personal ai and
just seeing what it, how itcomes back.
Suman Kanuganti (38:48):
Yeah, what I will do is um, I will uh have my
gates uh answer that.
So let me copy where you aretexting.
I will share my screen and Iwill.
I won't scroll because I Icannot show all the personas
that I have.
Uh, some of them are sensitive,so let's see, click share
button.
Okay, good, on the screen here.
(39:08):
If I scroll up and scroll down,I will have multiple different
personas across the company.
That are close to 60 of themFor this exercise.
I'm gonna go into Gates.
Gates is my I-I-C-O-O and Ijust asked a question about the
team, so that way I don't haveto show my history in here.
But what I will do is I'llsimply paste that question here
(39:31):
and send it to Gates.
Gates is trained on a lot of mycompany strategy-related
activities, so I'm going to lethim speak now.
Kevin Kerner (39:43):
Yeah, yeah, do it.
Suman Kanuganti (39:45):
Leadership enhancement through perfect
memory.
Having perfect conversationalrecall through ai would
undoubtedly make one a betterleader beyond just exposing
repetition patterns.
Here's why pattern recognitionthe ability to analyze
conversation patterns helpsidentify both strengths and
areas for improvement inleadership.
Communication consistency check.
Perfect recall ensuresmessaging remains aligned across
(40:07):
different that's good, refineddecision-making frameworks.
Even if repetition is exposed,it often serves a purpose in
leadership, reinforcing keymessages and ensuring alignment.
(40:29):
The key is using that awarenessto communicate more effectively.
This aligns with our vision atPersonal AI, where we're
building technology thatenhances human capabilities
rather than simply pointing outlimitations.
Our focus on memory,consistency and proven accuracy
makes this possible at scale.
Would you like me to elaborateon?
Kevin Kerner (40:49):
That's really good .
Even if repetition is exposed,it often serves a purpose in
leadership.
Suman Kanuganti (40:57):
So most of the things that otherwise I would
talk about and you can kind ofyou know based on our own
podcast, speaks a lot about howI talk on the perfect memory.
The interesting thing in hereis, if you look at this, you
know personal score 77.
Personal score right.
That is an indication of howspecific or in-depth you know
(41:20):
that response is as it comes tothe way I speak and the way you
know I talk about personal AI.
Kevin Kerner (41:27):
Yeah, that's cool and it really.
The other thing that's reallycool is the voice.
Using the voice of the, you cantrain it on the voice of the
person, which is really makes itmore it.
Just, I don't know it.
Just to me it makes it morepersonal.
Suman Kanuganti (41:40):
Yeah, well, you know, I'm a phone call guy and
when I drive I call my people,when my people are busy, I call
my AI and then have a great time.
It actually helps quite a bit.
I, you know, simply talkthrough the things that are on
top of my mind.
Yeah, it's quite insightful attimes.
(42:00):
Yeah.
Kevin Kerner (42:01):
Yeah, I'm glad someone else does that.
Now I do it with my kids.
When we're in the car we'llpull up ChatGPT and we'll just
ask it all kinds of stuff invoice mode.
Now you do that with personalAI and you will have a history
(42:23):
as well, as it will startshaping the entirety of you.
It just seems like it gives yousuch more perspective than you
could get.
Well, you may not be able toget those conversations that
happen all the time with others,and then you can just dial up
these personas.
Pretty amazing, simon,unbelievable.
I could go on on this stuff.
I can't wait to see what youguys do next.
I'm sure others will want tokeep up with what you're doing.
How should people get a hold ofyou, interact with your company
(42:45):
?
What's the best way to get totalk to you?
Suman Kanuganti (42:48):
That's easy.
Personalai will be the firstone on.
You know, Google search or anyother search.
My email is the letter S atpersonalai, so if you want to
write to me directly, that's thebest way.
Kevin Kerner (43:02):
Awesome.
Can you interact?
Do you expose any of the AIs toexternally?
Do you have any AIs out therethat people can talk to?
Suman Kanuganti (43:10):
There are some on personal AI Okay, and most of
our work is very much ultraprivate.
So, it's less external usecases but a lot internal use
cases, but we try to expose someof them just for people to play
around yeah, you're not tryingto build character dot AI or
(43:32):
anything silly no, no, in fact,you know, in 2023 there was
almost like a competitivelandscape between inflection AI,
character AI and personal AI.
But our approach has been kindof starts, you know, starts from
the people, necessarily.
You know, simply talk to theprominent figures, right.
(43:53):
Our focus was more centeredaround, like, the individual who
is actually getting on personalAI.
So there was confusion.
It's like, oh, this ischaracter AI.
No, it's not.
It's not character AI.
Kevin Kerner (44:03):
Yeah, yeah, and you're on the right path.
I wouldn't.
Yeah, it's characterized awhole different, whole different
ball of wax.
But yeah, it's, Suman, I'm sohappy that you joined that we,
that I reached out and youjoined us.
I really can't thank you enough.
This has been awesome and I hadfun.
Suman Kanuganti (44:22):
Kevin thanks a lot for doing this.
Thank you so much.
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
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
The Clay Travis and Buck Sexton Show
The Clay Travis and Buck Sexton Show. Clay Travis and Buck Sexton tackle the biggest stories in news, politics and current events with intelligence and humor. From the border crisis, to the madness of cancel culture and far-left missteps, Clay and Buck guide listeners through the latest headlines and hot topics with fun and entertaining conversations and opinions.