Scaling Secure AI: Turning Enterprise Data into Impact (Behind The Build with Tharani Part 1)

Today's episode is particularly exciting

because we're diving into how to get

because we're diving into how to get started with AI implementation as a

started with AI implementation as a mature company. Whether you are a CEO or

mature company. Whether you are a CEO or a developer who is curious about real

a developer who is curious about real valuable applications of AI within your

valuable applications of AI within your platform. Today's conversation will give

platform. Today's conversation will give you practical insights of how you can

you practical insights of how you can get started. Today we are interviewing

get started. Today we are interviewing Atomic's very own developer who has been

Atomic's very own developer who has been working with an established company in

working with an established company in the education tech space. We're excited

the education tech space. We're excited to have Taran Yan Seam talk through her

to have Taran Yan Seam talk through her experience of helping large organization

experience of helping large organization figure out their AI strategy, validate a

figure out their AI strategy, validate a few ideas and help them develop a road

few ideas and help them develop a road map for success. All right, thanks T for

map for success. All right, thanks T for joining us today to talk through your

joining us today to talk through your experiences behind the build. So um

experiences behind the build. So um let's get started with a brief intro

let's get started with a brief intro about yourself of what you have been

about yourself of what you have been doing and that will kind of get us to

doing and that will kind of get us to know you a little bit better and then we

know you a little bit better and then we can dive into the rest of the story.

can dive into the rest of the story. Yeah. So thank you so much for having me

Yeah. So thank you so much for having me and I'm really excited to be here to

and I'm really excited to be here to share my story and some insights about

share my story and some insights about what I'm working with. So I'm in the

what I'm working with. So I'm in the tech industry for around 6 years and I

tech industry for around 6 years and I started as a backend developer and then

started as a backend developer and then moved towards fullstack development and

moved towards fullstack development and 23 September I joined atomic build and I

23 September I joined atomic build and I really had a lot of chances to work

really had a lot of chances to work around with different projects in

around with different projects in different domains and uh from the start

different domains and uh from the start of this year we have started working

of this year we have started working with AI and it's really exciting and

with AI and it's really exciting and super interesting to the clients as well

super interesting to the clients as well as the project. Yeah, it's a bit about

as the project. Yeah, it's a bit about my background.

my background. Awesome. So, how did you get into tech

Awesome. So, how did you get into tech in the first place?

in the first place? Okay. So, when I was a kid or when I was

Okay. So, when I was a kid or when I was in school, I don't have any plan. I

in school, I don't have any plan. I didn't have any plan to uh have a job in

didn't have any plan to uh have a job in the tech industry. But uh I was very uh

the tech industry. But uh I was very uh keen in chemistry and my plan was to

keen in chemistry and my plan was to become a chemical engineer and

become a chemical engineer and unfortunately life had other plans. So I

unfortunately life had other plans. So I ended up selected into uh a really good

ended up selected into uh a really good university in an IT uh degree. So yeah,

university in an IT uh degree. So yeah, it's changed me a lot about a lot in my

it's changed me a lot about a lot in my life too and starting from that I

life too and starting from that I started love tech. Yeah.

started love tech. Yeah. Awesome. I find most people's careers

Awesome. I find most people's careers there's a group that has like a very

there's a group that has like a very strong plan from their childhood and

strong plan from their childhood and they kind of carried it all the way and

they kind of carried it all the way and then the other 50% kind of find a

then the other 50% kind of find a pathway that fits their lifestyle and

pathway that fits their lifestyle and just kind of go with it. it gives you

just kind of go with it. it gives you the most uh interesting adventures. I

the most uh interesting adventures. I think mine was very similar to that. So

think mine was very similar to that. So how did you transition into AI AI

how did you transition into AI AI focused development all of that stuff

focused development all of that stuff that you're working on now? So when I

that you're working on now? So when I was a kid, I have some interest in AI

was a kid, I have some interest in AI because one of my cousin brother was

because one of my cousin brother was working in an AI industry and I was

working in an AI industry and I was inspired by him a lot by the work he did

inspired by him a lot by the work he did at that time and also in my school days

at that time and also in my school days and also in the university. I have a

and also in the university. I have a habit of reading a lot of blog posts and

habit of reading a lot of blog posts and articles related to AI and also I wrote

articles related to AI and also I wrote some medium articles especially on topic

some medium articles especially on topic like Sophia about the humanoid robot and

like Sophia about the humanoid robot and about digital twin also neural network.

about digital twin also neural network. So deep in my mind I have this interest

So deep in my mind I have this interest to work with AI and fortunately we got

to work with AI and fortunately we got the chance to work with the AI projects

the chance to work with the AI projects in atomic build. Yeah. Was your pathway

in atomic build. Yeah. Was your pathway into building AI tech easy relatively

into building AI tech easy relatively easy or was it challenging when you got

easy or was it challenging when you got started?

started? It's not easy because um instead of

It's not easy because um instead of reading the AI related things I started

reading the AI related things I started by doing. So we really have we had a

by doing. So we really have we had a really interesting project at the start.

really interesting project at the start. So it's kind of an data extraction from

So it's kind of an data extraction from the invoice. So it's the start of my AI

the invoice. So it's the start of my AI journey. So with my c-orker we solution

journey. So with my c-orker we solution this uh we implemented this solution by

this uh we implemented this solution by using some AI technology. So from that

using some AI technology. So from that onwards I started learning the other

onwards I started learning the other things related to AI like the new

things related to AI like the new frameworks the technologies related to

frameworks the technologies related to AI and all um trained my uh social

AI and all um trained my uh social medias like LinkedIn and Twitter to

medias like LinkedIn and Twitter to serve more AI content for me. So I build

serve more AI content for me. So I build everything around me like AI things.

everything around me like AI things. That's a very clever way to force

That's a very clever way to force yourself to kind of be

yourself to kind of be exploring

exploring that area. Right. So, let's go into what

that area. Right. So, let's go into what your current project exciting work that

your current project exciting work that you're doing. Now, before we go into it,

you're doing. Now, before we go into it, I'll give a bit of a background based on

I'll give a bit of a background based on what I understand and then and then we

what I understand and then and then we can dive into the nitty-g gritties of

can dive into the nitty-g gritties of how you got to kind of get started. So,

how you got to kind of get started. So, you're working with like a mature

you're working with like a mature midsize a tech company that's been

midsize a tech company that's been around for a while, a very they have

around for a while, a very they have very large data set. So their ideal sort

very large data set. So their ideal sort of company to actually leverage AI

of company to actually leverage AI because of the wealth of information

because of the wealth of information they have. Um and then they reached out

they have. Um and then they reached out to atomic to support them with their AI

to atomic to support them with their AI exploration. So tell us a bit more how

exploration. So tell us a bit more how did this how did you get started when a

did this how did you get started when a company that big with a large platform

company that big with a large platform comes out to you and and ask these asks

comes out to you and and ask these asks you to support them? How does one get

you to support them? How does one get started in that in that kind of a phase?

started in that in that kind of a phase? Maybe get us get us from like the pre

Maybe get us get us from like the pre beginning very beginning when you get

beginning very beginning when you get exposed like how did you get behind the

exposed like how did you get behind the build and understand how you build the

build and understand how you build the knowledge and the tribal knowledge of

knowledge and the tribal knowledge of that company itself because you need to

that company itself because you need to know the foundation. You can't just go

know the foundation. You can't just go in and start throwing ideas in front of

in and start throwing ideas in front of the in front of the team and go yeah

the in front of the team and go yeah let's go get building because that's not

let's go get building because that's not enough context. You haven't had enough

enough context. You haven't had enough context. So how did you do that? So this

context. So how did you do that? So this company has a large featurerich platform

company has a large featurerich platform and they have a lot of sensitive

and they have a lot of sensitive information related to people below 18.

information related to people below 18. So we spend a few weeks I can say few

So we spend a few weeks I can say few days to deep dive into their system to

days to deep dive into their system to explore their features and

explore their features and functionalities and what's going on and

functionalities and what's going on and then we moved into the AI. So we found

then we moved into the AI. So we found some features that can really be

some features that can really be improved by using the AI. So we didn't

improved by using the AI. So we didn't forceful we we didn't forcefully use the

forceful we we didn't forcefully use the AI just for the sake of it but we found

AI just for the sake of it but we found some really interesting areas that can

some really interesting areas that can help that that AI can help to improve

help that that AI can help to improve those. So we found those kind of things

those. So we found those kind of things and we came up with around 15 ideas

and we came up with around 15 ideas initially and we had some discussions

initially and we had some discussions around those ideas.

around those ideas. Amazing. So you took several days to few

Amazing. So you took several days to few weeks understanding the product as a

weeks understanding the product as a user on the front end level. So you're

user on the front end level. So you're kind of going through the whole

kind of going through the whole experience and and then also I guess on

experience and and then also I guess on the technical front as well. You're

the technical front as well. You're understanding how the data is set up,

understanding how the data is set up, how's the architecture set up. So you

how's the architecture set up. So you have a good understanding of how

have a good understanding of how everything works. So you build the

everything works. So you build the tribal knowledge, you build the business

tribal knowledge, you build the business sort of knowledge of where they're

sort of knowledge of where they're headed, where they're going, what are

headed, where they're going, what are their pain points during the whole

their pain points during the whole experience. And because you have fresh

experience. And because you have fresh eyes too because you're new to the

eyes too because you're new to the platform compared to a company that's

platform compared to a company that's been building for years, let's say their

been building for years, let's say their employees or staff that has been around

employees or staff that has been around for a long time, you have fresh eyes

for a long time, you have fresh eyes because you go in, you understand, okay,

because you go in, you understand, okay, this is where the gaps are because

this is where the gaps are because you're looking at how the entire product

you're looking at how the entire product is working brand new, completely new to

is working brand new, completely new to this whole space. So you get ideas,

this whole space. So you get ideas, you're exploring, you're understanding

you're exploring, you're understanding how this thing works. And then you start

how this thing works. And then you start going and looking at, okay, where can

going and looking at, okay, where can you leverage AI to like add true value,

you leverage AI to like add true value, not just surface level AI tech like kind

not just surface level AI tech like kind of wave a wand and then kind of say that

of wave a wand and then kind of say that this is AI limitation. You're looking at

this is AI limitation. You're looking at true value that you can drive because

true value that you can drive because that needs to align with the business

that needs to align with the business values as well.

values as well. Awesome. So you came up with 15 ideas by

Awesome. So you came up with 15 ideas by going through this exercise and then

going through this exercise and then from there um what was some of the

from there um what was some of the exercises that you did to kind of define

exercises that you did to kind of define okay the out of these 15 ideas like

okay the out of these 15 ideas like what's valuable what's not are you

what's valuable what's not are you aligning with the stakeholders the the

aligning with the stakeholders the the leadership team and kind of reviewing

leadership team and kind of reviewing these things what what are you doing

these things what what are you doing there? Yeah. So for the 15 ideas, we

there? Yeah. So for the 15 ideas, we just grabbed some summary of what the

just grabbed some summary of what the features would be and we presented those

features would be and we presented those ideas to the internal team of that

ideas to the internal team of that specific company and we narrowed down

specific company and we narrowed down the ideas from five top ideas. So we had

the ideas from five top ideas. So we had several discussions and meetings on

several discussions and meetings on what's the real value for the business

what's the real value for the business as well as for the users by implementing

as well as for the users by implementing those ideas and also we prioritize the

those ideas and also we prioritize the complexity and the usability of those

complexity and the usability of those ideas. So once we did that and also we

ideas. So once we did that and also we included several people from their team

included several people from their team from product and design and from

from product and design and from engineering and from UIUX

engineering and from UIUX team. So we included all all kind of

team. So we included all all kind of different people from that company to

different people from that company to get the ideas of separate different

get the ideas of separate different people and finally we came up with five

people and finally we came up with five different or five topmost ideas that

different or five topmost ideas that stood up the list.

stood up the list. Okay. So you stopped around shared the

Okay. So you stopped around shared the ideas with the entire e- leadership team

ideas with the entire e- leadership team members

members uh from different disciplines and then

uh from different disciplines and then got their feedback so that you now

got their feedback so that you now mature your ideas that you have uh kind

mature your ideas that you have uh kind of initiated and then out of the 15 or

of initiated and then out of the 15 or 20 ideas you've now selected five are

20 ideas you've now selected five are the most valuable for the business. What

the most valuable for the business. What happens next? What do you do?

happens next? What do you do? Okay. So after that one we prioritized

Okay. So after that one we prioritized the ideas based on the complexity and

the ideas based on the complexity and also the implementation level the ease

also the implementation level the ease of implementation and we prepared the

of implementation and we prepared the product requirement documents a basic

product requirement documents a basic document with explaining what are the

document with explaining what are the features and what's the technologies

features and what's the technologies we'll be using and the scope of it and

we'll be using and the scope of it and the success criteria the success goals

the success criteria the success goals of each of that. So we prepared that

of each of that. So we prepared that one. So after that we also prepared the

one. So after that we also prepared the proof of concept documents because for

proof of concept documents because for the proof of concept we don't need to

the proof of concept we don't need to show everything that is in included in

show everything that is in included in those ideas. So just we prepared the

those ideas. So just we prepared the document with some major features in

document with some major features in each of the each of the

each of the each of the so out of the five that you have defined

so out of the five that you have defined you're now developing more sort of

you're now developing more sort of details to it. So this is like detailed

details to it. So this is like detailed requirements that you're building. uh

requirements that you're building. uh are you collaborating with the

are you collaborating with the leadership team also as you're kind of

leadership team also as you're kind of figuring this out?

figuring this out? Yeah, because they are use using a kind

Yeah, because they are use using a kind of software so some limited software. So

of software so some limited software. So we collaborated with their internal team

we collaborated with their internal team to get approval about the software

to get approval about the software because this is related to AI and also

because this is related to AI and also we are dealing with sensitive

we are dealing with sensitive information. So we have to be more

information. So we have to be more focused on how what the information we

focused on how what the information we are sending to AI and what's the output

are sending to AI and what's the output and input we are getting and we are

and input we are getting and we are sending to AI.

sending to AI. What sort of

What sort of we Yes.

we Yes. Yeah. So we put a lot of guardrails and

Yeah. So we put a lot of guardrails and we always consider about the personal

we always consider about the personal identifiable

identifiable information the PII and also we when

information the PII and also we when selecting the technologies we always

selecting the technologies we always care about the security of the data and

care about the security of the data and how to sec safeguard the sensitity of

how to sec safeguard the sensitity of the data we are sending to AI. So we uh

the data we are sending to AI. So we uh implement so we think about a lot of

implement so we think about a lot of technologies and methods to safeguard

technologies and methods to safeguard the uh data as well.

the uh data as well. Okay. Amazing. So part of that sort of

Okay. Amazing. So part of that sort of documentation and exploration is like a

documentation and exploration is like a lot of lot goes into just the not just

lot of lot goes into just the not just the scope of the work but it's like the

the scope of the work but it's like the architecture side the risks that are

architecture side the risks that are part of that implementation itself and

part of that implementation itself and how you're actually mitigating this and

how you're actually mitigating this and you're working with the team to make

you're working with the team to make sure that they are aligned and the

sure that they are aligned and the expectations are in terms of what the

expectations are in terms of what the what the outputs are going to look like.

what the outputs are going to look like. What goes into defining the proof of

What goes into defining the proof of concept from there? So are you choosing

concept from there? So are you choosing a certain number of features only? What

a certain number of features only? What goes thought process choosing those

goes thought process choosing those feature sets? itself. Why do you choose

feature sets? itself. Why do you choose those features in the first place?

those features in the first place? Okay, so for each of the ideas there is

Okay, so for each of the ideas there is a prioritize goal, right? There there's

a prioritize goal, right? There there's a goal. So we selected the features that

a goal. So we selected the features that can uh help us to attain that goal. So

can uh help us to attain that goal. So if you take an example, if you take the

if you take an example, if you take the cost something related to course planner

cost something related to course planner creation. So we are uh having a goal of

creation. So we are uh having a goal of extracting this number of causes from

extracting this number of causes from the provided cost list. So we set a goal

the provided cost list. So we set a goal for each of the uh proof of concept and

for each of the uh proof of concept and also along with that we tested the

also along with that we tested the technologies of frameworks we'll be

technologies of frameworks we'll be using for the actual system. So we there

using for the actual system. So we there are a lot of AI tools and technologies

are a lot of AI tools and technologies currently. So we uh read around a lot of

currently. So we uh read around a lot of blog post and articles to find the

blog post and articles to find the suitable uh technologies for the system

suitable uh technologies for the system and we also did a lot of test by using

and we also did a lot of test by using separate LLM models and separate

separate LLM models and separate frameworks and finally we came to a

frameworks and finally we came to a solid foundation for each of the ideas.

solid foundation for each of the ideas. So the proof of concept is really to

So the proof of concept is really to iron out if your initial like hypothesis

iron out if your initial like hypothesis is right and it able to give the quality

is right and it able to give the quality result that you want in the first place.

result that you want in the first place. So you're you're evaluating a lot of

So you're you're evaluating a lot of these things through the proof of

these things through the proof of concept to make sure that does make

concept to make sure that does make sense before you go into a production

sense before you go into a production build. That's really for proof of

build. That's really for proof of concept. How long does that have it

concept. How long does that have it happen? what what goes into like doing

happen? what what goes into like doing that multi-month project or is this like

that multi-month project or is this like how do you do the proof of concept

how do you do the proof of concept itself?

itself? So most probably we narrow down the

So most probably we narrow down the proof of concept it's estimation within

proof of concept it's estimation within 3 to four weeks because we only chose

3 to four weeks because we only chose the top most priorities prioritized

the top most priorities prioritized features in each of the ideas. So we

features in each of the ideas. So we narrowed down for each of the features

narrowed down for each of the features like one or two weeks based on the

like one or two weeks based on the complexity of the feature and also we

complexity of the feature and also we provided the feedback or else we

provided the feedback or else we provided the results once every week so

provided the results once every week so that we can get a really good output or

that we can get a really good output or response from their team and we

response from their team and we reimplement the thing or like um

reimplement the thing or like um remodify the flow to achieve their

remodify the flow to achieve their expectations and all. So it's kind of a

expectations and all. So it's kind of a repetitive uh process. It consumed bit

repetitive uh process. It consumed bit of time but we finally we were able to

of time but we finally we were able to came to a solution that can help us to

came to a solution that can help us to make this idea in a solid background.

make this idea in a solid background. So from what you're saying it sounds

So from what you're saying it sounds like AI you can't just take it off the

like AI you can't just take it off the shelf and then and then implement it and

shelf and then and then implement it and say you get the results.