Subscribe
Share
Share
Embed
Welcome to The NeuralPod.
The 0-1 machine learning podcast.
Chatting with ML Leaders, Researchers and Engineers who've built models, systems and products 0-1.
Vivek: I can save a lot of
hours for me myself, like, , if
I had to build something.
Like c climate itself, , what I
built, which is an observability
dashboard for open claw, which
I built with open claw again.
so if I, if I did not have open
clock , it would've taken me, least
like, yeah, two to three weeks.
Chris: Everyone welcome back to the
Neuro Pod today I am joined by Vivic
Chand welcome to the podcast Vivic
Vivek: Thank you.
Happy to have be here.
Chris: Yeah it's great to have you
on I know we've done a little bit
of chatting recently and you are
currently a a tech leader@booking.com
and lots of experience building and
and scaling systems which we're gonna
talk to you about today But also you've
developed an AI agent observability tool
for open cloud which we're gonna chat
about I'm really excited about that And
then as always we'll have a bit of fun
discussing the the future predictions
So yeah Thank you for joining us let's
let's get into it dear would you just
like to start with with your uh journey
Vivi can give a quick introduction to
who you are all the way up to booking.com
Vivek: Yeah.
Yeah.
So.
Kind of like, did my
engineering back in 2011.
after that, kind of like,
worked in a bunch of startups.
I worked at a start called Recruiter
Box, joined there as a third engineer.
it was, it was in the HR tech, kind
of helping, recruitment, better.
so basically like, kind of building the
whole product from the beginning, seeing
the journey, multiple versions of it.
And then it, eventually got acquired,
to a, to a bigger firm, Z hr.
And then eventually I moved
on with that experience to.
To my own startup, in the real estate
space, to make rental housing experience,
much easier, for finding the landlords
and the tenants and their, overall
journey throughout their, housing period.
So that was about, well, bird, but
although kind of like built the
product and stuff, I probably did
not crack the distribution well.
but it was a good
learning for almost like.
10 to 12 months.
then I moved on to, class Clap.
So Class Clap was, onto education tech.
it was, onto improving the
quality of education inside the
classrooms, making personalized
education scalable, across, India.
Pretty much like the first two
years kind of built more of a
MVP or like a pilot program.
yeah, kind of experimenting in,
in a couple of schools and then,
expanding it from, across India.
so that's something
I'm, I'm super proud of.
so after.
I felt, an urge for me to kind of
like work at a, at a really large
company, with all, everything
that I learned mostly in startups.
so@booking.com.
I've been kind of like helping on
the, on the customer support tech,
for the past, two and a half years.
Kind of like leading the whole,
customer support, AI initiative,
helping, build the entire, pipeline
to kind of like give high experience
for the customer support agents to,
To support travelers across the world.
kind of like giving them the entire
context of what their, problem is so
that they can solve the problem faster.
So use of ai, wherever possible to
make, yeah, the experience of travelers.
Great.
So that's, probably my journey
with, so far with booking.
And, that's where like I, I got
hooked with this AI so much.
so my LinkedIn feed is always full
of, All the updates coming from,
everything new, like there is clock
coming with something new and then
like, Gemini coming something new.
And then, open clock came in and
probably I got hooked a bit late.
probably it happened in from December,
but I got probably hooked, end of Jan.
so I started doing, OpenCL
and yeah, that kind of like
changed a lot of things for me.
Chris: Nice Yeah We'll we'll get into
that Let let's save let's save that So
I think thank you for the introduction
really clear you know you you built
production systems across ad tech hr tech
now travel I think that you know there's
a lot of people that are maybe leaving
big tech maybe not from booking.com
but going into startups and you know
they may have been working on large scale
engineering systems their their whole
career you know what what are the common
threads you've seen in building production
systems from you know smaller startups
to to uh la large engineering systems
like booking or or or even differences
Vivek: I mean, so basically like,
end of the day, like building any
of these prediction systems, it's,
it's, it's, it's easy to get started.
I mean today's, or like it's easy
to integrate with a, with a large
language model or anything to
build AI onto anything, but to
really bring out the ROI in it, it.
It, it's important to evaluate how, what's
the quality of whatever it's predicting.
it should be part of the system
wherein like, you, you make a
prediction and you also evaluate the
quality of it, with the eval loops.
So, I mean, the teams or the
companies that master, building
these eval loops really well
can, really make the best out of,
The whole ai, trend that's happening.
a lot of companies, yeah, I mean like
are still trying to struggle to find out
how to do this evals well, really well.
I mean, if, if they can
really, figure this out, yeah.
I mean that is the deal breaker.
Chris: Yeah exactly And you know I
think coming onto an engineer being
an engineer in the current market as
maybe a graduate you know if you think
about when you as a graduate it's prob
probably completely different now in
terms of being a junior engineer with
all the AI tools and co-generation tools
that are out there What what what advice
could you give to someone junior or
if you had someone junior joining your
team to operate in the current market
Vivek: I mean, it's important
you try out every tool out there.
Like, I mean, at least get hooked
to clo clo start building things.
Don't, don't think of like, yeah,
somebody has to give you the
requirements to, on what to build.
Like start building
whatever you wish you had.
if you, I mean, today it's, it's possible
to build such applications, on top of
like, say open Claw or any of these
new AI frameworks that are coming in.
So it's important you kind of like, be.
Hands-on with, all of these things,
new frameworks that are coming in and
the new coding tools, I guess that's
what is most of the companies are
looking for, expertise in this skill.
if you're able to kind of be up
to date because, like upskilling
yourself, like with the newest thing,
that, that would, that would be a,
yeah, a great thing to focus on.
Chris: Okay thank thank you for that Let
let's talk about C Cloy which is an AI
agent observability for for open claw
before we get into Clo c Cloy I guess
let's start with open claw you know I'm
sure everybody listening has heard of
it by now but let let's start with with
why you think uh observability is is
important for for a tool like open cloth
Vivek: so until now, probably like,
we kind of like had applications,
where it was about like.
there are, there were a bunch of services,
onto which, it used to make API calls or
like stream data from one node to another.
so we had Grafana for them, like we
had SLOs, everything built around it.
so that dashboards and like
we would, we would be able
to observe what is happening.
But now we are in a world where, Where
we are giving superpowers, I mean, we
are getting in the world where, the AI
agents have become, like, have more powers
and, they can get onto your computer.
They, once you install it, they
can do everything that you can do.
so.
I mean, the moment you give so much
of superpowers, it's important.
You need to understand, what
is really happening in there.
because, it's, it's no more synchronous.
It, it is, so many things
can happen in parallel.
There is like execution.
There is, fetching, there
is, running of new commands.
So it's important that like an
observability, High absorbability
on these kind of AI agents is
important, particularly to, for
the security concerns that we have.
Like, I mean, the moment you give full
access, I mean, if it, if it goes and
deletes something in your, in your
laptop or on your, in your database or
like, it, it goes and posts something
on your Twitter or in your LinkedIn,
you, you want to be careful of that.
And, and also that like, It
can pretty much do anything,
with, with, with that thing.
So that's the reason, like, having
a full visibility on what all it can
do, is, is but very much important.
If you want to also kind of like, yeah,
improve your application, to, to not
burn tokens, and, and just be focused
on what it's supposed to do, then
you need to have this observability,
Which will, which will give
you, which will tell you, which
will lacked as a feedback.
So there is one is, as I
mentioned, evaluation on whatever
your predictions are doing.
How good is the quality of that evaluation
and how observable is your platform?
Like, only when you can
clearly see what it is doing.
Then, your product, whatever you
are building, will be kind of like,
will do, will work as you expected.
so gone are the days
where, things were, broken.
either like one of the button is broken,
but now, it's, it makes a wrong decision.
And, you get a wrong answer.
So that is what is, is, is the new broken.
So you need to know when it's
making this wrong decision
with, with observability tools.
Chris: Nice And yeah I know we tried
offline just before the podcast You've
installed open Claw and you're using it
daily And I think to your point at the
start of the podcast you know you go
on LinkedIn and you see some of the use
cases and I do wonder what what's true
and what's not What what do you think
is are some of the good early use cases
for for Open Chlor and someone using it
Vivek: Yes, I mean, for me personally,
it helps me in, coding itself.
I mean, like for anybody who, who has
been using Cloud code, I would strongly
recommend to try Open Cloud, because, it,
it lets you have this long-term context.
so.
Gone are the days where I used
to open clock card and explain
or give the context of what the
feature was or what the project was.
I need to just chat on my telegram,
and I can say after two days I can
just continue on that conversation.
The.
The AI has the context and it can
continue building the features.
It knows how to deploy my code.
It knows, how to run
tests, everything, right?
Like, so this is something, was
not there with cloud code at
least, or with the, with the ones.
I mean like now I.
I, I'm, I'm aware that like the cloud
code is also, introducing something,
similar where like you have a long-term
memory and can access remotely.
So there are going to be more
players here, coming in with
the, with a similar, offering.
But, it's important for us to, yeah.
kind of like make the best use of it.
Like whoever is best right now.
right now it's open cloud.
so yeah.
So that's a good part.
Chris: Yeah exactly And uh out
of interest would how much more
productive would you say it has
made you or how much more efficient
Vivek: I mean, I can, I can do, I can
save a lot of hours for me myself,
like, if I had to build something.
Like c climate itself, what I built, which
is an observability dashboard for open
claw, which I built with open claw again.
so if I, if I did not have open clock
and I had to build clock code with
clock code, it would've taken me,
least like, yeah, two to three weeks.
what I was able to build with open
cloud, with the, with the long-term
context it had was, Pretty much, yeah,
over a weekend to like three to four
days, I was able to build like, a more,
kind of like usable product that I
can, I can open source and kind of like
share it with others to give it a try.
So.
It did help me reduce the
time and my cognitive load.
I did not have to, put a lot of cognitive
stress on like, what to explain to plot
code or any of the AI coding tools, but
rather, put my burden on, open cloud
and it was able to take it from there.
Chris: That's really interesting Let
let's talk about telemetry then We've
we've all I'm sure a lot of us have
seen some of the horror stories online
you know quite a well-known safety
researcher for example had their inbox
deleted by Claude Bot And I think that
Having an agent like this without any
guardrails is is quite frightening to
a lot of people you know do you wanna
just explain to people what telemetry is
Vivek: Yes.
so.
Kind of works on top of,
or sit next to Open Cloud.
open is your, is is your AI agent, that,
that you install on your laptop or, or
on your VPS or anything, wherever you
want to play around, like Mac Mini.
it, it currently kind of like listens to,
or like views all the files that, open
Claw is, logging into and helps you see,
or visualize what Open Claw is doing,
so that you can in parallel, see what it
is thinking, what files it's accessing.
What subagents, it's spawning
how much tokens it's burning.
Everything.
You get a visible, dashboard.
You send a message on telegram, like,
just ask it like what's your name?
so you can see in, telemetry.
How it's telling your name.
I mean, it's, if I ask what's
my name, like it's gonna tell me
Vwe, but how is it gonna find out?
Right?
Like, so that's, I'm, I'm, I'm a
curious engineer, to understand how
it's really finding out telemetry tells.
That it's gonna look into the context,
that it had, about the user md and, it's
gonna fetch that thing and, print it
back or send a message back on Telegram.
So you get to see everything
that is happening, when you chat
with your telegram, live with.
So that's the kind of
transparency it brings in.
and also when you ask it to, do a deep
research on something, something about
like, yeah, I mean some AI frameworks
or anything about like, about your,
your travel plans or sort of things.
So you can see actually like, open, I mean
open club making, even like open browsers.
Clicking on a bunch of things,
you can see the actions in
telemetry, of what all it's doing.
So that's the kind of visibility,
telemetry is bringing in.
And it's, it's open source.
it's, anybody can just like install it.
it works across, Mac, Linux, and Windows.
it's just one command be installed,
cometry, and you get it running.
Chris: Nice Yeah I I think we
spoke offline You've already
had 28,000 downloads in a
Vivek: Yeah.
Chris: short period of time
so congratulations on that
Vivek: Yeah.
Thank you.
Chris: where did you get the idea from
I know it's relatively new technology
and you've you know you've written
moved really quickly to develop
this tool what spawned the idea
Vivek: so I, I got to
know about open cloud.
A bit late, as I mentioned,
like, probably, end of Jan.
immediately I got to know.
I mean, I, I also heard that like,
yeah, don't install it on your personal
laptop or on your machine or anything.
I, I wanted to give it a try.
I tried it, I was really like amazed by.
so I was building a site
project called IC Voice.
it was a, it was a Sanskrit
learning app, so I was struggling
to build it with, plot core.
I thought like, let me, give it
on the hands of open claw, and
it was able to build super quick.
But I got a point where, when I
sent a message, I don't get a reply
for a long time and I don't know
whether it's doing something or not.
so probably I ensure on Jan 28th,
by fifth fourth, I realized,
okay, this is not happening.
I need some visibility tool.
And then I immediately started
questioning open clock.
What are you doing?
Like, can you help me
visualize what you're doing?
And then it said, I'm
doing all these things.
can you build a dashboard for me?
And it started building
a dashboard, itself.
and then, I, I got a local host,
server running, and then eventually
I asked more features on top of it.
And, Eventually, it, it outgrew
and, it became, commentary.
So that's how, that's how it got started.
It, it, it all completely was
built by open client itself.
Chris: su super cool okay And what what
what are the future plans for for Cloud
Tree A anything else that you're going
to be developing into the platform
Vivek: Yes.
Yes.
so the one thing that, I've heard from a
lot of people who are using telemetry is
that like, so they want this, The current
version that we have, open source version,
works on a, on a local area network.
so if you install it, it,
it works only in your land.
if you want to, kind of
like monitor it remotely.
I personally want that, like, I
don't want to, Like, be on the same
land and access my C telemetry.
I want it to be accessible remotely.
So that's one of the pain that
I have and most of them have.
So we are building a managed version like,
where like, c telemetry can emit events
to cloud and can access it remotely.
so that's, the managed
version that's coming in soon.
And, also an Android and an iOS app, where
I can, I can check from my phone from
anywhere, what my Open Claw is doing, and
also in a Mac app where like, yeah, I can
access it from anywhere, what Open Claw is
doing and also planning to, bring in more,
Security monitoring, in, in terms of
like, since open clock can do a lot of
things and since geometry is sitting next
to open clock and can view everything
it's doing, so, it is possible, to have
a layer where, if, if open clock tries to
get prompt injected, or we kind of like
figure out it's getting compromised, then.
We can alert to the user that
something, something wrong is happening.
Or if it's trying to delete everything
from your file system or all of these,
they can easily kind of like flag it.
so these are the features, that I'm
kind of like thinking, of building.
so already I have some, contributors,
within the, since it's an open
source project, so everything
is going to remain open source.
the only thing is that like the
hosted version, It's gonna be, a
small, few for, for a few small
dollars, , so that it can be accessed
from anywhere in the world as well.
Chris: Amazing Well yeah I'll include
links Tolo in the podcast links if
anyone's interested in downloading it
and getting involved you can download
it there is is there anything else
you'd like to add on Onlo Tree before
we move on Cool Okay Well yeah the the
last little part I think discussing the
future you know when we first started to
podcast Vivid we used to do predictions
Two years stand the Line But I think who
knows what will happen in AI in two years
It's moving so quickly So in terms of
you know six months ahead How AI agents
are moving where where do you see the
observability space moving for example
and what's still missing because as these
huge benefits of having these agents but
it also carries huge risk and I guess
as more corporations adopt this kind of
technology what what do you think is kind
of mi missing for for observability there
Vivek: So what I see is, there are more
open source, frameworks or projects
are gonna come in, like, like open
cloud, and there is already nano cloud.
There is Pico Cloud, which
already has some 20 K Yeah.
15 K Ktop stars.
So, and these, Frameworks are
gonna run on tiny hardwares, like
Raspberry Pie or even $10 hardware.
And, they're gonna be everywhere,
maybe, maybe six months to one year.
and it's important for us to know
what exactly is happening inside them.
I mean, at least as, as a creator of it,
like to know what it is, really doing.
It, it, it could be part of your
CCTV cameras or it could be some
kind of like a, any, any device,
like Alexa or something like that.
something similar.
so it's important for, for the creators
to know what it's really doing.
So it's important to have this kind
of observability and, how I see
like eventually, things moving on.
Is that like.
As these more AI agents are coming
in, it is, it is, it is going to
be more of managing these AI agents
well, to ensure we don't, yeah.
Which tokens, tokens
should be spent, in a.
In a well thought way rather
than, yeah, just getting burned.
It need to have this full visibility,
of what is really happening.
and eventually you will
manage more agents.
once you have this kind of observability,
you can manage more agents, like you can
have open claw that manages a bunch of.
Open cloud agents or with, with this kind
of observability like telemetry, where
you can delegate more things, to AI and
that further delegates to more AI agents.
So the imagination is, is what you need.
probably, if you can imagine
it, you can build it.
So that's the world we are getting into.
Chris: how do you see the wider
space playing out over the next
year what are your thoughts there
Vivek: Yes.
so as I mentioned, like it's,
it's gonna be, more, more things
getting automated and probably
like, it's, it's, it's, it's time.
Like you are building a
product, you need to build for.
Two users, right?
Like one is human user
and also for agents.
So agents are going to come to
your site, come to your product,
they're gonna do things, on behalf
of another agent or another user.
So that's one primary thing to
think of, when you are building
products, yes, build for both the
users, and this is gonna happen.
Yeah, I mean, you need to also think
like how your services are, kind
of like, monitored, because, it's,
it's, you might have thought only
it's like human users, but no, it's,
it's, it's going to be a lot of AI
that's new to have that observability
monitoring set up and probably
rate limits, if you don't want to.
Burn, yeah, your, your CPU cycles.
so those are things that, , to
keep in mind, when, yeah, for,
for the coming months, I'd say.
Chris: Okay And that fi final question
Vivi what what what tools I know you've
mentioned obviously cloud barometry and
and cloud code sorry open claw what AI
tools are you using in your everyday
life Stay productive Are there any other
niche or small tools that you live by
Vivek: yeah, I mean, so currently,
like I'm, I'm fully hooked to the whole
open cloud thing, but I'm, I'm, I'm
always like kind of like interested in
trying out new things like perplexity
computer, or the plot cowork.
so these tools.
Keep improving every day, and,
so that I don't have to do a
lot of manual setting up things.
so I, I keep myself trying out
new things that are coming in.
but yeah, right now it's, it's perplexity
computer and, clot cowork, that I'm
kind of like looking into as well.
in parallel to open.
Chris: Nice Yeah We're we're currently
looking at cloud cowork and looks
like a great tool So that wraps up the
podcast for today Vi thank you so much
for joining joining us and and sharing
your knowledge and discussing CLO and
and Open Law We'll we'll include the
links to to the downloads and where
people can support the open source
project So yeah Thank you for your time
Vivek: Yeah.
Thank you.
Thank you so much for having me here.
Bye.
Chris: You're welcome