YouTube Transcript:
AI-powered impact: Vertex AI for startups
Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.
Share:
Video Transcript
View:
[Music]
Well, thank you everyone. Welcome. How's
everybody doing? Are we in the in that
moment before or actually after lunch
now and we're like, okay, we launched
now. We see it and we're going to relax.
We're we're going to try to keep the
energy up, guys. So, this is a great
session. So, first of all, welcome to
Next. Hope you're having a great time
here. My name is Ilana Kines and I lead
the customer engineering team for North
America startups at Google Cloud. So we
have a startup hub downstairs in the
expo floor. If you have any questions
about the startups, please come talk to
us and meet us there. We be thrilled to
have you there. Uh but today this
session and these amazing founders are
going to help us understand some very
specific applications, use cases and
innovation that they are building uh
using Vert.Ex AI. So for those of you in
the audience that know the product or
are interested in learning more about
the product, this is a great session for
you. We're going to have a strategies,
lessons learned and some specific uh
capabilities that these founders have
used. So with that said, let me set the
stage and welcome our founders. So I'm
very happy and very grateful that you
are all here. Um I'm going to start with
Tete Tetia. He is the CEO and founder of
a company called Prompt AI and they
provide a visual intelligence platform.
So Tete has been in this space for many
many years and what they do is take all
the information that they capture
through visual devices and then act upon
them and the information that they can
gather. So thank you Teta for being here
with us today. Thanks for having me
here. Of course. Uh and then Akil Gupta
and I should say Akil welcome and thank
you so much for such a big flight. He
comes from India so great founder from
India market leading uh application
there. His company is called
nobroker.com and what they do is they
disrupt the real estate industry.
Basically providing a way to think about
removing the middleman from the real
estate transactions and do automated
property matching and detect fraud as
well. So we'll hear more about that in a
second. Thank you so much for making the
trip and being here today. Wonderful
being here. Thank you. Thank you. And
and then last but not least, so hi
Ahmed, welcome here. CEO and founder as
well of a company called Resemble AI. So
many of you may be familiar with them.
They do fantastic innovation in
generating voices and automating voices
with different models. So think about
all the fantastic spaces about cloning
your voice. Anyone has tried that? So
please do come to this company. Uh
congratulations also on your launch of
rapid voice cloning 2.0 just in February
I think of this year. Yeah. Thank you.
Appreciate it. Happy to be here. Great.
Thank you so much. Well with that said
let let's deep dive on this thing. So
why don't we start with you? So hey on
that one. So I think a lot of people in
the audience may be thinking okay you're
in a different industries all of you. So
what dropped your decision to use for
instance Vert.ex AI to build your
applications to build your solutions
because I'm sure there were things that
were out there that you were looking at.
So what dropped your your decision to
use Vertex? Um so we started resemble uh
five and a half years ago at this point.
So five and a half years ago there
weren't many things actually out there.
So actually Vert.Ex text didn't exist at
all at that point. But um uh we were
developing our own models um for
generating voices and um one of uh we
tried a lot of things to really get
scale in terms of the number of amount
of compute we need uh because we had
this interesting problem where every
customer every user would come in and
they would be building their own voice
model, right? Um so we had this weird
like one user has many model
relationship and uh what we really
needed to do needed to do is find a way
to like scale up compute really quickly.
Um you know we tried other services as
well for this uh solution. Uh but we
landed on at that time was ML engine uh
what was called ML engine at that time
which is now called Vert.Ex AI. Um but
effectively we found it to be super
scalable in terms of us having the
ability to train models very quickly and
um uh there were layers that we actually
ended up building on top of at that time
ML engine which is now vertex which are
now just incorporated right into vertex.
So uh you know we've kind of seen the
evolution and the growth of uh uh the
product and obviously um compute has
become a huge concern uh scalability and
uh availability of the compute has
become a concern as well. So uh for us
just having that compute available and
having horizontal scalability for
training purposes is what really got us
into the product in the first place.
That is great to hear and I know
scalability is a big thing just across
the board for for all the founders here.
So uh let me let me ask ail or or tete
do you find that been a similar uh point
of decision point for you to use vertex
AI? Yes, that's that's one of the
consideration which has really helped uh
even no broker and the convo and what uh
the product what we have uh because it
makes it easier earlier if you go back 5
10 years back and we we are a 11 year
old company so he says that he didn't
know about vortex 5 and a half years
back 11 year back there was nothing so
if you had to do something uh in terms
of whether you want to leverage AI ML to
build models or create something it used
to take a lot of time and that time has
drastically come down. It's like you can
now do it in few minutes to few hours
and there are a lot of foundation models
lot of models which are available with
vortex which can easily seamlessly do
the basic task for you. So that has
definitely improved. So and that's a
great point right because we have
scalability and we have different models
as well that you coming into the picture
and then you can decide among them. So
that's great and and t do you have a
similar experience too on the models or
the scalability side? Absolutely. uh I
mean we are in the phys visual physical
AI space. So in there like there's no
like a uniform model like you know
Gemini for voice or or for text for
example. So we have to train and deploy
a lot of different models for different
purposes and I think that's why Vert.Ex
X AI is is really really helpful because
you know it's every couple of weeks we
might have to address uh you know a new
application a new use case that might
mean like we have to tweak the existing
models or you know train a new one just
specifically for that. So we find vert V
vertex AI uh to be really really useful
for scalability. Nice. So I love this
fact and and I love the fact that all
these are good points for us to know
about Vert.ex AI. Let let me try to see
also on the other side for instance when
you were going through your journeys on
Vert.ex and learning what you could do
going from like nothing before to this
to ML ML and then Vert.ex AI any
specific lessons learned that your
yourselves or your teams actually went
through that may have been a little bit
hard at the beginning that were much
easier or just things that were pivotal
moments in that journey.
Sure. Yeah. So basically if you if you
see uh no broker we are a uh world's
largest broker free real estate
platform. So we save close to a billion
dollar of brokerage in India every year
and it's been 11 years. So we close more
than 70 18,000 80,000 properties every
month. So I'm talking about 2,000
properties being closed on the platform
uh or 200 300 properties being closed uh
as we speak. So now that needs lot of uh
uh smartness onto the platform and we
don't have any field on uh field uh
force on the ground. So identifying
brokers so that happens basis and
brokers brokers are called agents here
or the property consultant. So there's a
very very different phenomena with which
these people behave and they uh work
identifying their uh insights their
signals to when people are uploading
pictures. So what we do we tell owners
okay you please upload your pictures uh
via WhatsApp and when somebody sends you
the photograph for silly reason they'll
just select 10 15 uh photographs
together and one of the photograph can
be a good morning or a happy birthday uh
image and you really don't want those
images to be available on your platform
correct so there has to be a scrutiny
which happens so I'm talking about uh
1617 we had built-in model uh with
Google and that time obviously vortex
object was not there with the object
identification where we started
identifying what all objects are there
in the images with that we used to
classify whether this image is for a
house dining hall kitchen bedroom and
then we used to accept or uh reject the
images. So multiple use cases and to do
this particular use case it took us few
months at that point of time and
rebuilding the model training it tuning
it and there was no AI it was machine
learning at that point of time the
beauty about our industry is that with
every 2 years or 3 years you'll have a
new buzzword and people start using that
particular thing but we were building
machine learning models but now with
Gemini and with Vortex uh the models
which are available that thing has
become so seamless for us we had built
our own model. We had deployed it. We
were incurring a cost on that particular
thing. Now it's a SAS model. We just
pass on the image and we tell uh we ask
the model whether this is a image of a
property or not. And not only it tells
me whether the image of the property, it
can also beautify the image and give it
back to me. So those are the seamless
things which have uh definitely help
helped us in uh with Vort.ex AI. Thank
you. I think uh the the lessons that
we've learned over time for Vert.Ex have
been uh how well it fits into the other
ecosystem of uh
uh solutions that exist on Google cloud.
Um so I think Vert.Ex itself has
broadened quite significantly over time.
Um obviously we were using to train
models initially. So being able to train
a model, evaluate the model on how
successful it is. If it's like for
example like Gil mentioned for object
detection, uh if you were training
models right now, having the ability to
evaluate and figure out how accurate
your model is. um and you know combining
with other tools to kind of make that
work whether they're Google related or
whether they're not like you know we we
use a product called weights and biases
to kind of make sure that the product
that we're or the model that we're
creating um is you know evaluating
correctly is actually performing
correctly there's no regressions um but
also in terms of like you know we we
have this continuous stream of
foundation models that we're creating
and then being able to scale on vertex
AI from there uh has been phenomenal and
in terms of data storage you have you
know a bunch array of options, anything
from like hyperdisk to just plain Google
cloud storage and how that all kind of
integrates together um into this one
product where I think a lot of the
development team is very happy to kind
of work with a bunch of Google tools but
everything kind of exists in one product
or another which is actually a a great
thing I I guess for the audience to know
right so part of the power of vertx AI
is that integration integrated vertical
stack and solutioning for you to make it
easier for your teams and your
developers to actually go from
prototyping experimentation on new
things and then actually deploying into
production for your customers. So Tet
how have has these particular function
functionalities from Vert.Ex text help
your in your case right using the models
that are provided using the APIs using
the all the integration capabilities
absolutely so uh prompti is uh uh still
in this early days so we've been
operating for about 18 months and that's
kind of an interesting experience so I
kind I want to share that from that
perspective so when you're early stage
company many times you have to do a lot
of trials and errors and iterating a lot
through these use cases and processes so
um and I was uh I had my PhD at UC
Berkeley before and that was very
different from doing academic research
where you want to get everything right
and many times in startup you really
need to get a direction right first do
these iterations and try to be as
efficient and fast as possible so
sometimes we make engineering compromise
and sometimes we look for off-the-shelf
solutions first before we decide to
delve into that and I think vertex AI
has been tremendously helpful in that
sense because you know we can put
together a solution very quickly and and
this was not un unimaginable a couple
years back then and now we can just put
them together and very be very fast in
like actually go to market and deploy it
uh to a either it's a group of uh test
users or the our entire user base and
sort of iterate from there and if we
realize okay we got to do model training
we'll do that afterwards but then we can
still deploy them on vertx for example
and Gemini for example has has been
really really uh transformative because
now it's sort of you have this uh
general almost like a computer where
like it understands instructions it
understands natural languages um so we
are able to uh build very high level uh
applications very quickly using these
these APIs which is great and I I think
you all pointed out to to one thing
which is your teams like vertex they
like developer uh developing with the
platform but let's get a little more
specific on that. So how long if you can
give us a sense right for the audience
especially um does it take your teams to
go from let's say an experimentation
phase or like a trial phase for new
parts of your product your solutions to
then having uh the model train and
everything ready to go and then
deploying. So can you give us a sense of
like what that journey looks like? How
long does it take? Oh I can share a
actually an interesting story. So we
were uh building up this pet feature. So
it's you know visual AI helps you with
anything kids pets of course. So we were
I want to see my dog in Absolutely.
Yeah. We got to recognize them. I
understand like this your dog is a
neighbor's dog and if your dog is doing
thing any any sort of uh anything it's
not supposed to do right. Um or anything
just that's interesting and fun. So we
were sort of like you know uh
implementing this feature and our uh uh
design product officer uh he moved
pretty fast but he was taking his time.
It was like, oh, the engineering team
was going to, you know, take a while to
implement this thing, especially, you
know, we're implementing this as a a
full scale feature that's going to be
pushed out to everybody. And, uh, he
estimated it was going to take us six
weeks to get it done and we got it done
within three. Three weeks. Three weeks.
Oh, so that's 50% cut of the original
time frame that you guys expected.
Absolutely. and he usually, you know, uh
says like I tend to bend space time in
the company uh because whenever they say
it's going to take eight weeks, I like
how about four? Let's let's work that
out. How about four? Um and this time it
was genuinely surprising like we we got
it done within such a short period of
time. That is pretty amazing. So it's
and great productivity uh gains for you
as as your development team. Absolutely.
And meet this deadline. So that's fine.
You just need to be careful, right? Next
time he will tell you two weeks
expecting it to be ready one week. I'm
always greedy. There you go. All right.
So, well, actually he and so any any
similar experiences hopefully. Yeah, I
can go I can go first. Um, so we have a
we have a model that we've uh deployed
into production that can detect deep
fakes. So can detect images, audio, and
video if they're if they're AI generated
or not. Um whether that's from you know
a Google openai doesn't matter who's
producing this. except it's open source
models etc. Uh a key part of that is
actually like uh a curation of synthetic
data. Um so uh we actually use uh we're
continuously like upgrading these models
and this is almost an automated
functionality now. So the idea is that
um we have this crawler that goes out uh
and that is observing different GitHub
repositories and a hugging face etc. and
it's trying to figure out if there are
new commits or new models that are being
that are being published. It scrapes
data, puts them into a cloud storage
bucket, um, uh, puts them into a Excel
sheet or Google sheets. At this point,
this is probably most untechical part of
this entire process. We'll remove that
Excel we're recording. And effectively
what ends up happening is uh, as soon as
some QA person says like, oh, this is
actually valid. Here's a here's a data
set or here's a model that our model has
not seen before. And the regression test
shows that the current model has low
coverage of this. um it'll immediately
trigger a model training on Vert.Ex
through the to a training platform a
custom job that's what they call it um
and effectively train a model
immediately to get that coverage. So for
example in the last week or so there
have been week and a half there have
been um three models that have come out
like Gemini 2.5 now supports image
generation OpenAI supports image
generation through chatbt and Midjourney
came out with V7 and all of those models
even though Midjourney came out on
Monday or Tuesday uh if you go upload a
picture from that product today it'll
tell you that's fake. Um, and the reason
for that is because it quickly gathers
data, does a regression test, and it
immediately kind of fires off a custom
training job to kind of train that model
and get that coverage. So, it's uh it's
kind of built in a way that kind of puts
all the pieces together. Well, it's
saving you but also saving all your
users a lot of time, right? Yeah. The
users expect like, you know, if there's
a new model that comes out um uh an
image generator, video, etc. Uh our
users expect coverage almost
immediately, right? Otherwise, if you
have like a firewall or a spam filter
that can only catch spam like a month
ago, then it's not very useful because
your attacks are are enhancing almost
every day or every week. That's true.
And and what an amazing use of the
product too, right? So detecting deep
fakes and all that. So that's great. And
using that for productivity gains, but
also time to service, time to market,
right? That's pretty pretty important,
pretty good for you too. And with those
all those properties I killed right and
all the services that you provide how
does that work for you in terms of the
time I think uh there are uh hello yeah
so if you see at no broker uh we don't
only help people find houses or buy
houses. We work in the all facads of the
property which is like you may want to
get moving services. You may want to get
your uh rental agreement, sale deed,
property deeds, uh you want your house
cleaned, you want your house painting
done and if you see all these uh
services they have a touch point and
they need uh somebody to go visit your
house maybe to see how much of the area
has to be painted so that I can give you
the quotation for that particular thing.
when you are moving how much is the uh
quote how big is your house because
typically and I'm sure this happens
across the globe whenever somebody asks
you how much stuff you have to move you
will always say I have little but when
the truck comes which is supposed to
take the luggage and with the people who
are married you'll suddenly find so many
lofts which have sudden uh stuff coming
out and typically it overflows so for
that uh we used uh Gemini and uh some
beautiful applications what we have
done. Now what we tell our customers is
that take the new broker app and if you
are moving just roam around the house
with the video on and when you are
roaming around just open your wardrobes.
If you have the beds which has the
storage just show us how uh of the how
much of the stuff is there. Let us know
if that fridge has to be moved, this
sofa has to be moved, TV has to be moved
and then we calculate what is the cubic
capacity, what is needed to move this
particular house and uh what will be the
cost of moving that particular house.
Imagine earlier we were doing a
guesstimate which 60% of the time was
not working well because of the hidden
stuff which is there uh in form of toys
of your kids or maybe the old clothes
what you have all those things we are
able to do now so that's one and this
this phenomenal then second one is when
you do your lease agreements again uh
year uh typically in India it happens
after 11 12 months you had to ask all
the details on the form Now what we do
we just tell them whatever lease
agreement you have in whatever format
whatever language it has been written
just upload it we just scrape it we use
uh OCR we get all the details and we ask
three four information like what's the
new rent what's a new deposit just fill
in those details click confirm boom your
rental agreement is ready so all those
things which were taking like days and
which was uh earlier needed human
intervention to do this stuff all of it
we are able to do with AI now which is
Great. So, not just eliminating the
middleman, but all those uh potential
services of someone going to check on
what's the space require and the service
and how much we're going to cost and all
that. So, that's pretty impressive.
Thank you, Ailio, for for sharing that.
And actually, I could use some of those
services too.
Not in US yet. Not yet. Not yet. Not
yet. All right. So, um so now let's
think about you've been working
obviously with Vert.ex already for a
while and uh the product has evolved,
right? it was non-existent then it was
ML then it's vertex AI today and there
have been a lot of announcements um at
next this week right about vertex and
some of them are related to agents and
agent building and some of them are
related to new models so what I would
like to take the conversation now
is how are you looking at the future for
your companies and how some of these
announcements some of these new
developments advancements actually can
help you power those that new next layer
of innovation that you you are thinking
about for your companies. So if you can
let us know a little bit about that and
and that will give us a glimpse also of
where your industries are going too.
Ted. Yeah, I can go first. Um so uh
imagine the future where like these
spaces are are watched by AI so that we
don't have to spend hours watching these
videos and also this information coming
back to a centralized place and uh we're
just able to ask questions about what
had happened and uh the insights of of
what had happened and that means for
example the well that means the first
step is to understand environment right
visual understanding And after that it
has to be agentic because it needs to
connect uh the things that had happened
to intentions to uh what we as uh
operators or users, homeowners, business
owners, what they'd like to see. And
these things are all different.
Sometimes they are personal. What I want
for my home might be very different from
what you want uh for your home. And uh a
uh retail shop owner what they are
trying to uh uh get might be very
different from a hotel owner for
example. And these models will have to
and systems I would say have to be able
to understand intentions and work in a
way that that different people uh might
want very differently. And uh I think
for example uh vertex AI can be a a very
um important role in that. uh for
example reasoning capacity for for these
models and now they have to think and
step step by step laying out what they
have to do and now they have to go to
the uh prospective uh parts of the
system uh whether it's a storage they
might have to check some data in a
storage or they might to go have to go
into the database and and uh come up
with a a SQL query uh some keyword to
search for some information and on top
of it they need to synthesize this
information and then decide what to do
next and or stop and present that
information to people and sometimes it's
even a voice interface. So we're really
getting into the stage where computers
are getting really
sophisticated and uh and also like we
just have AI to automate a bunch of uh
the either the boring task or sometimes
it's just very heavy for human beings to
do. Yeah. And I love the fact that you
mentioned I think all these things that
you mentioned you started with saying
it's agentic right it's a lot of these
process flows that are going to be built
on top of that and vertex can give you
capabilities right agent SDK and the
agent builder and all those uh those
parts of the product so how is that um
helping you or potentially helping you
and let me go with a or so um are you
planning on using those are you already
in that journey of the agent u building
how agentic those solutions will be for
you uh any specific things that that you
can share with us? So uh so at the scale
of no broker where we have like close to
5,000 employees working for us and most
of them uh a big chunk of them or a
majority of them work in our customer
service department where they have to
touch base with the customer answer
their queries understand what they need
like I was talking about packers and
movers I was talking about cleaning
painting and all those things so then
but for a customerf facing company the
SOP is that you should have a consist
consistent great quality service which
is unbiased by the mood of your agent.
Correct? It should it should it should
not happen that I had a fight with my
wife tonight and or early in the morning
and I'm disgrunted on my customer and
I'm not happy to help that particular
customer and that had always been in my
mind that as we grow big how are we
going to solve that particular thing. So
we started building models very early.
So now what we do we have built a
platform called convoen.ai which is like
zen out of customer conversations. C uh
customers can be conversing with you on
a chat chatbot emails SMS WhatsApp and
on your call center. We take all those
conversations and India the beauty is uh
we talk in multiple languages. So we
have like 14 15 languages which are
actively used otherwise we have hundreds
of languages and people switch
languages. Uh so they'll be speaking in
English and suddenly Hindiad that's what
I did and it it comes very very
naturally to us. So none of the models
were able to solve that particular
problem. So we created our own ST models
and now once we had that particular
thing we were able to create agents like
agent assist where there is a virtual
agent who is sitting on top of our
platform and one of my call center
executive and he's he or she is talking
to the customers it can tell you what
exactly is the history of that customer
that okay she came to no broker platform
3 months back this is what she had or
maybe she has active service going on
she had sent you an email she's not
happy about something which is going on
this is what you need to tell so
basically uh the things like okay sir
can I put you on hold and then I'm going
back I'm going to search with my manager
all those things immediately goes off
now when you talk to the customer you
say okay hi this is what is happening I
see that you have a packer remover
movement and our partner has not reached
I have already put uh uh put a touch
with my partner and he or she may be
reaching uh in another 30 minutes so
that that levels up your experience uh
to a different level. Then after we did
that we realized that there are a lot of
task which don't even need human because
I feel as a human we should do something
which is non-mundane we should we should
be thinking we should be uh creating new
stuff we should be doing something smart
so then we created our own virtual
agents uh you can say humanoids which
can talk in Indian languages uh and
that's what I was talking with Zah also
like uh it can make a call to you it
will feel as If a human is speaking to
you and if let's say you have a property
visit scheduled it will just call you
and say hi I see or you have property
visit scheduled and are you coming or
not and then somebody says oh no I see
there is a traffic oh I also see that
there's a traffic okay so that means
that you'll be delayed by 45 minutes
that that is what Google map is showing
let me just reschedule the appointment
for you and I'll also inform the person
on the field who was supposed to be with
you on that particular visit so things
like that we have started automating and
that's where the agentic
theme has started coming into uh our
platform and because it was so beautiful
uh we have started selling it out as a
product uh to other companies also there
you go so another business revenue
stream there so that's good
congratulations on that hill and thank
you for sharing it so what I'm hearing
also is that it's not just the internal
experience that gets better with all
these new advancements but it's also the
experience for your customers of course
right so not just for the internal
developers that are using the platform
but also the end result So no, no
company can be successful until your
customer is happy. Absolutely. And I
love hearing that it's actually good for
you to use our technology on both sides.
So thank you for that. Um but so hi so
let me ask you because in your space
specifically right there's a lot of
innovation going on with models out
there from you from other companies
there's a lot of competition there's a
lot of innovation that we're bringing to
the table. So how are you navigating
through that and how do you see really
the future for resemble AI is going to
look like with your technology with the
help of of Google but also with things
that are going on out there that are
coming out. Yeah. So I'll answer this in
two ways. So um we're kind of lucky that
we develop models and our customers go
use those models and applications. So we
have a lot of insight and oversight as
to what applications are are um are very
useful uh and where where they're
creating an impact. Right? So we see
everything from like call automation and
I think uh these two gentlemen have
talked a lot about different automations
and different agents that are really
applicable and everyone here is probably
tired of hearing voice AI for the last
two days. Um so, uh one of the things
that's probably the most impactful in
resemble and you know we we've uh I've
actually like worked with a circle of
other founders to kind of implement this
in inside of companies and we're we're
really bullish on this actually is
um I I'm a firm believer that every
company should have one dedicated person
ideally a team but if you're if you're a
startup one dedicated person in just
exploring different agents and how they
could applicable as employees in your
company. Um, and that has like
tremendous benefits to the company and
it's now way easier than ever, right?
So, you can actually get employees that
could do programming, you know, there
there are literally software out there
if you if you wanted to get something
off the shelf. There's Devon, you know,
uh there's plenty of others. Uh there's
customer success uh products out there.
Um with a with a the real power here is
um every company like every human is
also slightly different from one
another. uh but the building blocks of
you know using Gemini using OpenAI using
different models to achieve different
tasks is a matter of plumbing work
together and then the core really
becomes how it works in your workflow
right so a lot of us um and I'm really
bullish on this is the most valuable AI
company the most valuable agent AI
company is probably Slack right now and
the reason is because every AI like
agent that your company will interact
with it's like an employee within Slack
so why would that be any different um So
having these agents being deployable,
it's like having uh staff that has
10-second SLAs's. No human staff member
can give you a 10-second SLA. Uh but an
AI agent can. And there's a lot of uh uh
a lot of uh great stuff happening within
these companies, including Resemble. You
know, we're deploying uh bots that are
effectively helping customer success,
internal, external. There's different
ones. Uh we're hooking them up to
different products. We we have a bot
that typically sits on our um on our
document page which helps people make
integrations because at a certain point
we're not going to write and maintain
SDKs for every single language. Uh it's
too much work for us. Uh but what we can
do is we can effectively have a have a
you know a chatbot that's geared solely
geared to understand our SDK and our
documentation and then the user can go
in and say oh I need to plug this into
Genesis or I need to plug this into
Unity. How do I do that? Right? And of
course, we're not going to write a guide
for every single integration, but this
thing can this thing can do it on the
spot on the fly. Um, so creating these
like agents, um, particularly
internally, which is kind of where I
have the focus right now, is it can pay
a lot of dividends and it helps your
company learn extremely quickly. So, uh,
I'm not sure what the audience makeup
is. If you own companies, you should
probably be doing this. If you don't,
then you should probably go to your boss
or manager and say like there should be
a team or a group of people dedicated to
just experimenting with agents just
internally making those workflows
better. that I love that idea and let's
actually quiz the audience. So just by
show of hands uh how many of you are
maybe already doing that creating agents
internally going to your managers and
saying hey we need to to do this for
some of those employee tasks and
functions that are very repetitive or
that are intelligent but could be better
done with AI right now. Showing of
hands. There you go. We're like 30% of
the room. 40. Yeah. Yeah. Just about.
There's a while to go. It's a lot to go.
Yeah. Exactly. But we're just starting
in that journey. So I think it's coming.
It's coming. So yeah. All right. No,
that that sounds great and thank you for
sharing that. Um All right. So since we
have also of course founders in the
room, one of the things that uh that I'm
sure they are probably thinking about
also is with your companies. You are in
different stages, right? So TE's
companies earlier on, you guys have been
uh for a few years already. So what's
next for your company? What are you
excited about for your company for your
next milestone?
Uh let me go first. So so if you talk
about no broker we are 11 year old
company the only prop tech unicorn in
India but given uh at this stage also we
are just present in six cities in India.
So we have a lot and lot of uh ground to
cover and uh with AI and with kind of
automations uh what we are able to do at
the uh and the rate at which technology
is changing uh I think companies will
become global. So it will be the
solutions uh which you'll be able to
create from one country and it will work
across the globe and that is something
which people keep asking me that uh when
exactly are you going to come uh to
different countries uh because in US
also you see that uh the amount of
interpretation cost is extremely high
and with lot of things happening uh on
the uh on the law side. So there's
opportunity for us uh there also but
right now we are focusing on India uh
very big opportunity uh with no broker
no broker services what we have and the
convoen which is uh our customer uh
intelligence uh AI product what we have
built so we'll focus on that great so
hopefully we'll see you soon too in the
US and then we'll be happy to move you
there you go all right thank you Tede
what's next for you absolutely so promi
was founded by a group of PhD students
and professor from from Berkeley. So all
of us uh have been working on computer
vision for for myself personally it's
been a decade and for uh one of my
colleagues he's been working on it for
more than three decades since early days
like um when he was at MIT. So we just
had this frustration back in the days of
like well we've been developing so many
different algorithms and research works
and but how come these cameras are still
dumb cameras like how come they are just
recording and I have to go back to it
and I have used a slide tiny little
slider and to look for what I'm trying
to get and they can't really talk to
each other they can't not uh really they
don't really understand any sort of
information and how come that we've been
doing so many years uh so so many years
of work in computer vision and they
still can't tell you whether your cat
has jumped onto the couch or not. It's
not supposed to be that hard. Okay. Um
so that's why we started and now we're
getting closer and closer. I think we're
we're at the down of of visual physical
AI. I mean like you heard the word
physical AI all the time, right? Robots
and drones and autonomous agents um
everywhere. But then you think about it
like who's going to watch them, right?
And you you got to deploy these cameras
everywhere and so that you make sure
that they're not they're not functioning
or or doing bad things. Um and I think
our our goal is to uh sort of have
computer to be able to do anything that
only requires a pair of eyes. If we just
need a human being to sit there and
watch, please do that with a computer
because um you know we humans can do
much more interesting things and we can
spend our time more efficiently. We can
spend our the time with with family uh
with friends and and focus on the work
that actually require our attention uh
rather than just like these tiny little
things. So that's why we're I'm really
excited about the future. I think these
uh technologies can transform how people
interact with the home with their with
their pets and their environment also
like how businesses function. It's going
to make us more secure uh feel more safe
and more connected. Great. Thank you.
And I I guess what I'm hearing from you
also is that there's of course software
solutions that you provide today and
maybe the hardware pieces are coming up
at some point too. Yeah, absolutely. And
I think a lot of these uh hardwares have
been really commoditized like 20 years
ago. I remember like a camera, nice
camera would cost uh at least hundreds
of dollars if not like thousands of
dollars and now they cost 20 bucks. You
can buy them from like anywhere almost.
Um and it's not hard to to manufacture
them either. So the reason that um many
people are still not buying them uh is
that they really don't find a use case
for that. It's like I buy a camera, I
put it there, so I forget about it and I
pay like cloud storage for that. uh and
now finally people are able to get some
usage out of it and I think it's just
going to drive this very positive cycle
where people keep buying more cameras
and as a result we discover more use
cases we try to automate them and they
become happier they buy more cameras
true and that that point they will need
your visual intelligence platform too so
absolutely which is great all right
thank you for that so hey so let me
close that this section with you in
terms of uh what's next for the company
and what are you excited about. Yeah,
there's a lot to be excited about. Um,
just to give you context, uh, 4 years
ago now, 2021, we're in 2025. Yeah,
that's four years ago. Yeah, time is a
blur. Uh, four years ago, uh, one of the
things that we, one of our customers
actually published a show on Netflix
called The Andy Warhol Diaries. Um, it
the entire narration in the Andy Warhol
Diaries, Andy Warhol of course passed
away uh, in the 70s or early 80s.
Um and every narration from him in that
documentary was completely AI generated.
Uh I called this the pre-hat era. Um and
uh that gave us an idea of mainstream
use of generative AI that was nominated
for four Emmys uh that show or that
documentary series. And um that got us
thinking or got me thinking a lot about
okay this piece of technology that you
know in 2021 four years ago is able to
reproduce something that a normal
consumer that's watching TV cannot tell
if it's AI or not anymore. And um if you
fast forward today you have this in
pretty much all the modalities. You can
go and obviously create gorgeous videos
with open uh with open AI or Google with
V2 now etc. Um but you can also go in
open source and do them. And I don't
think open source is slowing down. I
think open source is keeping ahead with
the pace of where uh the frontier models
are. So the thought really comes in when
we're creating these platforms
especially as resembles creating these
models and allowing you know millions of
users to use and create models
themselves is how do we do it in a safe
manner? How do we get people to not be
able to scrape a video off of YouTube of
Akil and effectively just, you know,
clone his voice, take his face, create a
version of him, you know, and that could
be extremely dangerous. Nobody will do
that. Somebody might do that. You know,
we've had people on YouTube that have
said like, "I found my voice being used
by a different channel." Uh we've had
people, you know, complain about I never
was on this ad. I never promoted this,
etc. You have politicians obviously. Um
and uh the the thing that we are really
bullish on now and that we really want
to have impact on to be honest. We hope
that the company plays some part in this
is the deployment of responsible and
safe AI. And those are not just meant by
guard rails, but you know, as I'm in the
Bay Area, Tete's in the Bay Area, the
the way we think in the Bay Area to be
honest is technology is the answer to
problems as well, right? Technology can
be solved by technology and not
necessarily policies, right? Um, and so
we've been, you know, building models
around watermarking. We've been building
models around detecting defakes. We open
source models around like, uh, speaker
identification and person
identification. Um but all of those are
kind of coming together and we're trying
to really wrangle around this this
foreseeable problem where you know early
in January this year 55% of the internet
according to a lot of research
researchers um was being created with
generative AI right uh and the
projection was by the end of 2026 that
90% of it would be created by generative
AI. I think by the end of 2025 with the
uh image and video models that are
coming out that are widely accessible on
your phones at this point, 90% is a
pretty pretty conservative
efer AI being used in content being
produced. Um so that really opens the
door for malicious users on the other
end that can also use that content. And
what we want to do is actually give
tools and give models to people and
companies that are deploying these
models to also offer ways to kind of
prevent um kind of the responsible or
encourage the responsible use and
prevent malicious use of those models.
So I think that's where a lot of my
attention and focus is going because I
think generative AI is out of the it's
out of the box. This these models are
going to improve. I have no doubt by the
end of the year it'll get faster,
better, higher fidelity. Um that that's
a given at this point. So the the
response is well what's the what's the
what's the uh what what's the counter to
what's uh what we're about to see here
happen in the world. True. And I'm so
glad that you mentioned that because the
general concept really when we think
about guard rails is like regulations
and policies and what can we do but you
mentioned something very specific which
is well technology can also regulate
technology. So that's an interesting
concept and I think a lot of companies
are actually looking into that because
policies regulations will not be able to
advance as fast as we need them to catch
up with what's happening in technology.
So that's an interesting concept of what
you brought brought today. So thank you
for that. And actually we're getting
close to closing the session. So I'm
going to say kind of like a rapid fire
oneliner. What would be your advice for
founders in the room who would like to
use Vert.ex AI in their solutions today?
And uh what advice can you give them?
Just oneliner very quick. Let's start
with Ted tip please. Yeah.
Um so speed is really important for
listed startups. You got to try your
best for that. Thank you. Yeah, I think
uh the same. So basically the rate at
which you can innovate uh with vortex
and anything else uh is uh extremely
fast. So things as I was mentioning 10
years back things which were taking
months few years back which was taking
days now is taking hours. So if you
think about a problem which you want to
solve uh you should be able to do a P of
that particular thing extremely fast to
know whether it's going to work or not
and that can define uh how fast you want
to work on a problem. Thank you for for
that. I hear just go to
aistudio.google.com and click all the
buttons and you'll learn everything
really quickly. That's a good one too.
Well, thank you so much for that. I hear
that um hopefully this session has been
useful for you. Um I have to say thank
you to all the founders obviously
everybody else in the room and if you
are not this is my commercial if you are
not familiar with Google for startup
cloud program please come talk to us
startups hub in the expo hall and uh
thank you for being here today. Thank
you for investing your time with that
with us today and thank you for
evaluating or using our technology
already. Thank you and have a great rest
of your day at the
PL. Thank you.
[Music]
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
Works with YouTube, Coursera, Udemy and more educational platforms
Get Instant Transcripts: Just Edit the Domain in Your Address Bar!
YouTube
←
→
↻
https://www.youtube.com/watch?v=UF8uR6Z6KLc
YoutubeToText
←
→
↻
https://youtubetotext.net/watch?v=UF8uR6Z6KLc