Hard Problems, Smart Solutions: The Newfire Podcast

Gordon Wong: Welcome to Hard Problems,
Smart Solutions, the Newfire Podcast,

where we explore the most complex
challenges and groundbreaking

solutions with industry leaders.

I'm Gordon Wong, VP of Data
AI at Newfire Global Partners,

and your host for this episode.

Today, I'm thrilled to welcome
Sandeep Dhamale, Director of

Engineering, Data and Intelligence
at the American Medical Association.

Sandeep has built his career leading
data and platform engineering teams.

At the AMA, he has spearheaded
transformative projects like Datalabs

GPT, a private LLM infrastructure.

and the AMA Intelligent Platform, which
brings a modern, reusable technical

infrastructure to AMA's data initiatives.

Beyond his roles at AMA, Sandeep
has also served as an advisor with

Newfire, bringing his expertise
to help drive innovation and

scalability for our clients.

Today, we'll explore how healthcare
organizations can adopt scalable

data solutions to improve
patient and operational outcomes.

Sandeep, welcome to the podcast.

Sandeep Dhamale: Thank you, Gordon.

Really excited to be here and I
look forward to a fun conversation.

Gordon Wong: Yeah, me too.

This is, these are typically a lot
of fun and I've been looking forward

to speaking to you for a while.

One of the things I wanted to
ask you about really is how

you got into this space, right?

You were originally in fintech.

Can you talk a bit about how
your career journey took you from

fintech to healthcare and how you
can, how your fintech experience

shaped your strategies for tackling
healthcare's unique data challenges?

Sandeep Dhamale: Yeah, great.

I mean, I consider myself very lucky to
have had a career journey that I had.

Basically, after I finished my master's in
computer science, right, I landed this job

at a global fintech firm called SunGard.

The job was great for my career.

It allowed me to expand my experience way
beyond software engineering skills, right?

It was a global development firm
with an emphasis on building

global market connectivity.

So I was really talking to
different various global exchanges.

So that allowed me to work across multiple
business lines and wear different hats,

which involved building large scale
platforms from scratch, modernizing legacy

platform onto these platforms, right?

So thinking about opportunities
for new product development,

new ways of accessibility that
comes from the modernization.

So, the point being, while I was building
my engineering skills and building teams

and I was, building departments, I was
also keeping track of customer problems,

driving the customer listening sessions.

And I was also lucky because of that
I got to travel internationally.

But one thing that really taught me
that as an early career engineer there

was the importance of shifting left on
the scale of product life cycle, right?

And really develop that product sense.

I think product sense is such a crucial
skill in all high performing engineering

teams that I've worked with that it
definitely needs to be underscored.

So coming to the experience, the
platforms we built there, especially

towards the later part of my career
in SunGard, were with derivatives

processing, but the focus was on high
throughput, low latency types of systems.

You're now thinking about in memory
database, how to avoid context switching

and from processes to processes to
get the best performance you can.

While we were doing that, I got to
build these platforms from scratch for

data processing and data connectivity.

And then eventually also build an
API store for our customers to make

it all available for integration.

So that was a great experience, and
now I'm thinking about those concepts

similarly in healthcare, right?

So security, compliance, handling
of sensitive data were all critical

in those environments, and those
principles translate directly

to healthcare as well, right?

When we think about financial privacy
and integrity, it's crucial we think

about those things even more when we
think about patient records, right?

I also think about Fintech, and even
other industries that I've looked

at in general, the place of pace
of innovation feels faster because

stakes are a little bit different.

Even with regulation, there are no
patient lives at stake there, right?

So pace definitely feels
and looks different.

I do feel that API-driven ecosystem
and modern infrastructure were

far more advanced in Fintech
before I got to the AMA.

And I kind of think that kind of led
me to think about how can I think

about scalability and flexibility
into some of the problems I'll be

solving here into healthcare, right?

Think about iterative value
creation think about unifying

data assets and accessibility.

Those things translate really
well into healthcare too, right?

So that's how I think about those things.

Gordon Wong: You underscored
a familiar distinction for me.

Like, frequently as engineers,
we talk about things like

velocity or total volume, real
time analytics and so on, right?

And so, as engineers, we might think
that fintech and healthcare are

different, but what I just heard
you say is that when you bubble out

to the problem, the larger problem,
they actually have a lot in common.

Is that right?

Sandeep Dhamale: Yep, that is correct.

And that's one thing, a cool thing about
being an engineer, you're able to look

at the commonalities across the problem
spaces and bring on thinking that really

makes you see the art of possible.

Because you, if you've solved a
problem a certain way, why can't

you solve it for this industry?

Those are the kind of things
that engineers always love.

And I, I have seen those commonalities
first hand for sure, especially when

it comes to, uh, solving for data
unifying all of those data assets to

leverage the value from that data.

Gordon Wong: Multiple years in the
fintech world, built some really cool

solutions, now you're in healthcare.

Healthcare data comes with unique
challenges, such as silos and compliance,

legacy systems, heterogeneous systems.

What are the foundational
hurdles you've encountered, and

how have you addressed them?

Sandeep Dhamale: Right.

So foundational hurdles, like some of
the things that you've mentioned is

healthcare is in a state what it is
because of various different reasons.

Of course compliance and speed
being one of the reasons, but

because of that the problem spaces
of legacy infrastructure exist

a lot more.

So data fragmentation has creeped in a
lot in my opinion because there are so

many different systems, they're siloed,
the data doesn't talk to each other.

That has been one common theme that
I've noticed across healthcare.

It's also a lot of build versus
buy challenges that come into

places and decision making that
is happening in a siloed manner.

So there's duplication of platforms data
being replicated across multiple places.

So there's that redundancy that's
happening which actually some of the

recent regulation like HIPAA and GDPR
are only putting a more spotlight on

like if you look at where is your right
to forget or if you go to look for those

records, it's not just in one place,
but it's in like 26 different places

across the enterprise that you learn.

And then, then Joel on some days
emailing spreadsheets of the same

data to somewhere and that only
exacerbates the problem, right?

So I think the hurdle was the
fragmentation and trying to think

about how can we organize this data
in a centralized place so that we

have a better control and better data,
unified data management in place.

So that was one of the first
ones that I think about.

I really also think about governance
frameworks and compliance for designs

is becoming very, very crucial.

But while I emphasize those, I also
want to emphasize flexibility for

growth and thinking about scalability
in mind are as important as well.

So yeah, I, I think the hurdles
are really data fragmentation and

interoperability and how do we
really get those all together.

Gordon Wong: You know, at Newfire, one
of the things I do with our clients

is I advise them to think about the
ROI of their data efforts, right?

You have to consider what are you going
to get out of this and what are the costs.

Now, in fintech, it's probably a little
bit easier to measure that, right, because

you're looking at financial returns.

Do you have any suggestions
on how to measure the ROI of

data efforts in healthcare?

Sandeep Dhamale: Sure.

You can look at it two
different ways, right?

Definitely start with your use case
in mind and what you're really what

personas you're trying to serve, because
that's going to determine your ROI

case, because in some cases it's a data
product that you're building, which is

going to have a real financial incentive.

And that's why you're building it.

In some cases, you're really thinking
about administrative use cases.

And you're really trying
to think about how.

How this problem will make everybody's
life easier and get the type of

compliance and processes that we need.

And ROI in, when I was building
infrastructure, we've done it both ways.

Uh, one is thinking about what is
our current total cost of ownership

of these siloed data systems and
how much are we really spending.

And two, thinking about when we get
to our target vision of a unified data

infrastructure— what is the cost of
running the business going to look like,

and how do we transition from A to B.

And there are always
so many wins to be had.

You're, you almost always certainly
decrease your cost and reduce your risk.

So that's that's a win with
cloud based architectures.

You're always guaranteed more
and more flexibility that kind of

prepares you for the future evolution.

And.

And then you realize that while you're
building this infrastructure, if you

attach a couple of data product or real
revenue generating use cases to back

it up, you really bake in the cost of
maintaining this infrastructure that

kind of gets offset by the, by those
revenue generating products that you've

included in the first cut of the draft
that you're thinking about the vision.

And, and it has now unlocked value
because you can build the product 3, 4,

5, 6 fairly quickly with reusing all of
the infrastructure that you've really

built in for first couple of things.

And then the ROI really scales from there.

So definitely when you're thinking about
it, have those principles in mind is what

I is what has benefited me and I think
should benefit our listeners as well.

Gordon Wong: We use a lot of
metaphors in our industry, right?

We talk about DAGs, directed graphs,
we talk about pipelines, we talk

about manufacturing moving data into
knowledge, but more and more I'm

thinking that the correct metaphor is
that this is actually a marketplace.

We have sources on one side and we
have consumers on the other side and

we're trying to enable transactions
and any good marketplace should take

as little of a cut as possible, right?

Sandeep Dhamale: Yes.

Gordon Wong: Right?

So does that, does that
metaphor hold in healthcare?

Do you think, do you find that valuable?

Sandeep Dhamale: I do actually.

The more and more I think about how to
bring value to the data, it's really to

try to think about how you're really going
to cater this data and what packaging

and what formats and how little or
how big of that package needs to be.

So you really have to
think about the personas.

One of the other thing that's
where data marketplaces are

becoming so useful in my mind.

And it also shifts to the
healthcare because I think the

problem is, not very different.

You're thinking about interoperability.

You're thinking about making it more
accessible and accessible is the is key

when you think about marketplace like
concepts is like you're coming to a place.

I always think of it like
a grocery store, right?

You go to a grocery store aisle you,
you can pick up a you, you know exactly

what you're doing in the grocery store.

You know where to find, Your vegetables,
where you want to find packaged

foods or where you want to find milk.

So there, it's all well organized,
where the dairy section is and all that.

You can go to that section, you pick
up a product of the shelf, you can

read all the labels, you know what
you're getting and how it was sourced,

etc, etc, what the ingredients are,
or the nutrition information, right?

It's readily available for you to make
your decisions right there and there.

So you don't feel very confused as to
what you're going to walk out with.

You're going to walk out with a
whole milk or a 2 percent milk,

those are the kind of things.

So I also think about data
marketplaces the same way.

If you have your data products
very well listed and what you're

getting is clear for the consumer

that's a win.

That's how you should think about
packaging your data products in a way that

a consumer comes to a data marketplace
that they're getting what they're needing.

And that's one metaphor.

And then the second metaphor also
that really resonated with me once

was especially thinking about privacy.

You take that grocery store and add
a pharmacy section to it, and now

you need to have a prescription to
get a particular product, right?

You're walking up to a pharmacy and
you're telling, I have a prescription.

So that's to me most, some of the data
assets in that marketplace could be

going that way because they're protected.

They are important and
with sensitive information.

So personas need to be going that way.

So I agree with you.

I think data products and the
marketplace's vision translates well,

even into the healthcare ecosystem
because we're trying to really bring

the same type of consumer experience to
healthcare that we've seen elsewhere.

Gordon Wong: This is turning
into an economics podcast.

Some of the concepts you brought up
earlier were definitely, you talked about,

basically fixed and variable costs, right?

And if costs are lower, you can cover, if
you have a baseline products that cover

most of the costs of the platform, you
can afford to bring additional products.

So that seems pretty key to scalability.

All right.

So I want to hear a little bit
more about some of the things

you specifically built, right?

Particularly, I to hear about
the AMA Intelligent Platform.

Can you tell me about that?

Sandeep Dhamale: it's going to be easy to
explain AMA Intelligent Platform because

of some of the constructs we talked about.

When I started at AMA in 2019 and
when we were looking at our future

of data products, one of the ways we
were thinking about how can we bake

in flexibility into an architecture
that can allow us to scale for future.

So some of the exact same principles
that I described previously

were were in consideration.

We wanted to build a technical reusable
infrastructure on, cloud basically

so that we can scale the way we want
to and add, keep adding capabilities.

So think of it as, as more like a modern
data platform initiative where you're

thinking about what is your unified
data strategy is going to be, where

you're going to land the data, what kind
of tools you're going to bring in to

process the data and create your gold
slash product repository, if you will

what that infrastructure looks like.

And then build capabilities on top of
it of creating different API endpoints,

creating different type of integration
technologies for external facing whether

they're external facing product consumers
or internal users even, but thinking

about data accessibility from the get go.

So it was a culmination of all of this.

So it's baked with technical reusable
infrastructure in the back end with

a digital front door on the front
end that allows our customers or

even internal audiences to interact
with a developer portal built in.

So you can you can look
at our API stack directly.

So sort of, gives you a marketplace,
like a construct, but makes makes it easy

for our customers to really interact with
our data and make it more accessible.

That's, that, that was the vision
that we've been going in with,

as we do more with this data, it
also creates a framework so for

customers to predict what, how it's
going to look like and feel like.

It's bringing some of the
standardization to access patterns.

So that way, when we launch a new data
product, it's just going to be a walk in

the park because they've done it already
once and it's going to look very similar.

And the, all of the constructs that
they did with the first integration

are going to just stand in there.

Right.

So that's how we think about
AMA Intelligent Platform.

So for what it's worth, for a lot
of our newer data initiatives,

they're all being facilitated by
this platform infrastructure now.

And that has given us
flexibility for thinking about

our data assets in the future.

Gordon Wong: Incredible.

So when you first, when the platform
was first conceived What were

some of the real world problems
that you were hoping to solve?

What, what kind of pain
were people facing?

Sandeep Dhamale: I will categorize
into two sections, if you will.

Uh, One is pains of how can we get our
infrastructure ready for innovation?

Because we've had a lot of
legacy systems in place.

And then the second is, of course,
the real customer voice, right?

Like we've been hearing you,
you ship in certain formats.

In a lot of cases, there were flat
files, you, we don't know how to

track change history of these data
assets or in some cases, like it's not

intuitive, the documentation around it.

And it takes me a long time to
understand your data once it's delivered.

So those were some of the key challenges.

When we were thinking about architecture,
of course, we wanted to make sure we

take a simpler approach of not trying
to replace everything at once, but

think about how can we bring a data
platform that can be fed with some

of the existing data assets that we
have to create an infrastructure to

think about how to solve problems.

And then from a customer listening
perspective, we were applying lenses

as to, OK, you need more metadata
along with our existing data assets.

So how can we generate
that metadata for you?

OK, you need modern delivery mechanisms.

In some cases, that just meant making
sure we have API delivery formats.

Some were really low hanging fruit.

Some were really thinking
about where this is going to go

from , from our perspective, as
well as our customers perspective.

One of the good things is
people really rely on AMS data.

It's been powering
healthcare ecosystem for

decades now, and we just want to make sure
we are good stewards of these data assets

and continue to give it to our healthcare
partners and customers in the right

time, right, format and, and make sure
that we continue to solve those challenges

that we've solved in previous era as well.

Gordon Wong: Now, I'm sure that every
project planned along the way was no

problem, and you hit every deadline.

But were there any challenges?

Sandeep Dhamale: Of course, there's
no software project that ever

throws a curveball at you, right?

I think there were, right?

But I think key to success in any
of these initiatives is making

sure you have an end in mind.

I always say be stubborn on your vision,
but be flexible on details and take,

think about how your iterations could
look different based on your priorities

and what personas so your
scope can be changed as you're

learning about the marketplace.

So think about what your milestones
and their definitions can be

can be and still add a value.

So definitely go incremental smaller
iterations and think about how can

you ship a value every iteration.

And then for your bigger
milestones, be flexible as you're

learning from the marketplace.

I can tell you, for example, there were
certain use cases that were not, that

we thought were high value early on.

And then to, to some of the market
research and even internal hurdles

that we thought, okay, it's going
to take us longer to launch this.

But we were able to pivot that
into a different type of package

and make sure make sure we still
deliver value to our customers.

But also our infrastructure
initiative doesn't get stalled

because you always need to make
sure your customers are happy.

And that kind of gives us the
impetus and funding, in fact, also

to make sure that you can self
sustain and self fund the platform

initiative that you've started here.

Gordon Wong: Would you be willing
to share some of your wins with us?

Feel free to brag.

Sandeep Dhamale: I really think the
biggest one is is organizational one

is taking everybody along on this,
and I think some of the builders

will relate to it is it's one
thing to make, know what is right.

And then you having a vision and
being able to also evangelize that

across it takes a village, right?

So you have to really be able
to talk to all the departments

and all the stakeholders.

And making sure that this this resonates
with everyone, and this is the most

important problem that you're trying
to, so the framing of this initiative

it took us a little bit I'm going to
say a lot of attritions to get it right,

but I think I feel very good about it
right now that we are on a right path.

Uh, so that's one.

From an infrastructure and technical
perspective, I think getting our

developer program stood up was one
of the highlights of this initiative.

So I'll give you a back story, right?

Like, we were building this
infrastructure for all the

reasons that I mentioned to you.

And then we had an initiative where we
wanted to make our CPT asset available

to early stage builders with an open
license kind of a thing, where they can

get access to our CPT content when they're
building their use cases and they've

not figured out their go to market, for
example, and they want to have access.

It was a perfect marriage because
we had an infrastructure in place.

All we had to do was to think about
what a CPT developer program could look

like, invite these builders onto the
AMA intelligent platform, and reuse the

same infrastructure that we have, right?

We have a developer portal, we have
an API store, and that we had built

for our and curated for our customers,
but we were able to repurpose

and launch a CPT developer
program in matter of weeks.

Like I, I think from conception
to reality, it was like four to

six weeks of project and boom, it
was launched and it's one of the

programs that has given us a community
that actually participates with us.

We were able to take
those connections forward,

we interacted with our conferences,
we have quarterly Dev Chats.

And they've been actually, uh, some of
the people who are working on cool use

cases with our content and data products.

So it's always useful to hear perspective
because industry is like you have matured.

Participants that are using your
data assets and there are some

new use cases you're seeing, like
how people are thinking about

what they can do with your data.

And having that voice baked
in has really helped us.

So I think that was one of the
coolest achievement of AMA Intelligent

Platform was to be able to launch a
CPD developer program, for example.

Yeah.

I think what's next is, is definitely
AI and generative AI on our minds.

I also think it's, it was a good test
when AI wave came along is some of the

investments that we'd started making in
this infrastructure stands true or not.

And I think it has, right?

If you think about, doing your AI or, more
importantly, generative AI now, right?

You need to have the infrastructure that
backs it and you need to be ready with

your data in a unified store and the right
formats and the right point of delivery.

And I think this initiative has
started to push us in that direction.

Some of the interesting things that
we're thinking about there has been

more about unstructured content that
have been sitting in PDFs and different

because a lot of our data assets were
also, one of our business, I'm going

to say, was really a book business
or a human oriented business, which

is now turning into a data business.

So that's a shift.

So we're taking some of those
assets and trying to trying to

really create RAG architectures or
graph databases and vector stores.

Those technologies, we were able to
onboard onto our data platform in

fairly a quick amount of time because
we had started landing all of this

into our data platform already.

This was just, this is the good part
about modern data platform that I

think is you can bring another piece of
technology into your stack fairly easily

and it's able to, like you're thinking
about your architecture all the time

is volume, velocity and variety, right?

So this variety of, the new variety
of data and how do you really bring

it to bring it to take advantage of
vector storage, for example, what was,

is something that we're working on, right?

Like you asked me what's next.

So this is definitely on our minds.

And there are some couple of use cases
that we can maybe talk about that

we're thinking in that space as well.

Yeah.

Gordon Wong: Yeah, well,
I'd love to hear those.

Sandeep Dhamale: Sure.

I mean, yeah with AI, when a couple of
years ago, I think it's coming up on

the second anniversary now when Chat
GPT launched we, I think early on, we

pretty much jumped in with thinking
about what use cases and what are the

framing that we want to put in place.

I think we were bucketing our use
cases into a everyday AI and a

transformative AI type of category.

With everyday I mean, can AI help
us automate some of our tasks, make

lives easy for our people give them
hours back in the day and solve some

of the search problems even that
we had with these assets, right?

So that's where we focused first is is
trying to train a model and create a

RAG infrastructure on our book business
oriented data asset to see whether we

can unlock some real value and create
an assistant for our internal staff

to ask questions and see if they can
find right documents and references.

So it's a combination of a search
problem and a a really easy chat based

interface to think if this intelligence
layer can give you the answers

right off the bat, because even this
internal team that I'm talking about

gets questions from different teams
and even external people on guidance.

So it is, the goal is to
accelerate their workflow.

So that's a pilot that we're running
right now with the hope that this is the

same infrastructure that we can actually
package intelligence into future and think

about what kind of intelligence offerings
we can make into future products, right?

So that's how that's an initiative
that we've been thinking about really.

Gordon Wong: Yeah, so keep leading
into the marketplace, right?

That's what it sounds like.

Sandeep Dhamale: That is Correct

Gordon Wong: One of the conversations
I have with our clients at

Newfire with I say, you know,
encourage 'em to think about ROI.

They get very tired of hearing
me say ROI over and over again.

But that's, that's really
kind of everything.

We have both the, the numerator,
the denominator, the value

we generate and the costs.

And I think personally I see a
tremendous amount of value in

using AI to lower those costs.

Lifting constraints, removing blockers.

What are some.

What are some areas of high friction
you see in delivering data products

that you think AI can help with?

Sandeep Dhamale: I think I've got a good
example for the question you're asking

I mentioned we, we were a book business
and we're trying to get into a data

business with that particular data asset.

First challenge is making sure you
can segment the data the right way.

Right?

You have to break the content down
into a different format altogether.

And then, you have to look for
common keywords, common tags,

like create metadata on the fly.

So,

giving this to humans,
to curate a new data

asset, reading through books, and
doing it like that, it's gonna be a

It's going to take you forever to do it.

I think what we're thinking about it is,
and we've done already some pilots with

this is, can you use large language model
infrastructure to prompt it properly to

create segments out of a big book pdf
and see what kind of output it generates.

Then take those sections and create
like a preview summary that you can

attach as a metadata tag into to that,
create keywords for that sections

and generate like 10 keywords each
for that section and create that.

And again, the goal is to accelerate
the stewardship of this new

data asset that we're creating.

So these are tools that we're
providing to our data teams or

our content teams specifically.

That who will be in charge of reviewing
this content and essentially stamping it.

So these are tools that are going
to assist these data owners.

To really look at what tags got generated,
how it's really creating sections of it.

So it's not like you're taking and
fully automating it, but you're really

making their lives easy and they are now
reviewers of this process rather than

taking a book content and try to break
it down themselves one by one and reading

through that's such a daunting task.

I think AI is going to make that much more
simpler and thinking about curation and

metadata creation process, that's one area
where you're gonna see your costs go down.

And especially these initiatives
now are generating value, right?

Like this was a value that
was sitting somewhere on a PDF

document that was not getting

uh, solved for now you have really
created a data offering out of it.

So you have an opportunity to go
to market with a new offering.

That's what we are seeing
with this kind of initiative.

So whatever total addressable market
space you're looking at with that kind

of content is definitely needs to be
factored into your ROI calculation.

That's one.

And then the second thing I'm going to
say is, because these data products are

giving are going to the new markets,
you're going to get more feedback.

For example for, in our case, a real
example, this was going into only

into the hands of a certain section of
people who actually read this content.

Now we're able to take these micro
Intelligence, if you will, and

we're thinking about whether we
can inject into newer workflows.

So that's like a newer market
you've you're thinking about now,

not just the previous market that
you were working with, right?

So I think you really have to be
able to re-imagine your domain and

where it's heading in the age of AI.

And then then think about
what possibilities exist and

which bets you want to take.

But yeah, there are going to be bets
that are going to be available to

our, to like the listeners of this
podcast, because I think once you

have data, you, the next layer you're
thinking about is intelligence, right?

So I think the AI plays the role in
simplifying your data processing as well

as thinking about new use cases where
your intelligence can be delivered.

Yeah.

Gordon Wong: As responsible stewards
of the platforms we build, I want

to take this opportunity to remind
us of two expressions, right?

One is knowledge is power.

And then the other one is with great
power comes great responsibility.

Sandeep Dhamale: Absolutely.

Gordon Wong: So how do you see AI helping
us address the privacy concerns in

healthcare and protecting our patients?

Sandeep Dhamale: Great
question again, right?

I I think two ways.

One, and especially, I think we mentioned
this in somewhere in the podcast, right?

When you're thinking about building
your data products, think with

your end users in mind and the
personas you're building with.

And they're going to be, once you
do your market research, you can

actually work with large language
models to frame those personas right.

And what kind of controls you
need in place, like defining

them itself can be done with the
help of large language models.

But I think, The daunting task of
automating the processes that you need to

have in place for each of these personas,
like what are those processes going to

be and then be able to create standard
operating procedures according around it.

It's always a time problem, right?

You don't have enough people to think
about think about these controls.

You can automate a lot of these
controls with with help of AI.

I think it's a problem.

And a balanced space because AI is going
to make accessibility more prevalent.

And then how can you also leverage
AI with security first mindset

to think about automating the
processes and the governance

framework around it is important.

I think some of the things that you
talked about data marketplaces and how

you can get your access streamlined
because of a data marketplace or that

kind of an access pattern where people
are coming to the same place to get

their data I think it still remains.

And these are the kind of platform
thinking and processes will continue

to make sure you have a governed
platform and not just an accessibility

that has now created a newer type
of problem in security space.

Gordon Wong: I like to find the
security people in whatever organization

I'm in and ask them to teach me
how to be their best customer.

Because otherwise, I'm just a walk,
I'm just a walking problem, right?

Sandeep Dhamale: Yeah.

Well, one thing, one thing that I think
especially in recent times, I've had a

lot more appreciation is probably, and we
did this in our recent all hands within

our engineering team is to start thinking
like everybody's a security engineer.

That's part of your job.

We've just tried to ingrain that
into our build teams mindset

is our centralized security team is only
so much and they are going to evangelize

the principles, but it's everybody's
responsibility and really thinking about

that at the design time is very important.

That also brings me to the
architecture point, right?

Like evolution, evolvability
versus maintainability.

And that's a, that's always a balance.

You want to have an architecture
that scales and evolves.

But it is also maintainable because
if it's not maintainable, your

security is going to be a nightmare.

And that's why we're trying to bring
a balance is it's your responsibility.

It's everybody's responsibility.

And we're trying to think about whether I,
I don't we've not done it, but we're now

going to start putting like a line into
job descriptions also to start emphasizing

this point more around security as
we think about further and further.

Yeah.

Gordon Wong: I'm going to take the
opportunity to jump on the soapbox.

I'm going to try to frame
my statement as a question.

But I personally have seen that it feels
like most organizations underinvest in the

maintainability of their platforms, right?

They focus on development costs,
but you don't really deliver value

until the thing, until whatever
you're building is in production.

Now, have you seen the same thing?

Sandeep Dhamale: Yes, all the time we
see this around and it's also, I, like I

was saying, it's a balance and sometimes
engineers have to be that voice of reason

because I know everybody, including our
stakeholders are excited to see the value

to market and time to market as the most
important thing and which is, which it is

in my mind and you go through phases in.

In my mind, you go through phases.

If you're trying to build an MVP
that you want to head to the market

quickly and see what it does, you
can think about you can think about

what your maintainability looks like.

But if you're making that decision, you
really have to make sure that you're

baking it as a part of consideration
as to what what it takes to really get

to the production grade product, if
you will and have that on your roadmap.

You don't just discount it.

You really make sure that it's out there.

In my mind, You don't really put
maintainability completely out of

the window from the beginning, you
actually bake it in, you do trade offs

and try to work with those trade offs
based on your time to market, but have

a roadmap, make sure it's out there
and everybody has seen it, that your

MVP can hit with X parameters, but to
get to the production grade, you have

to have X plus Y in place without that
you don't really go live is important.

And it's also allowing them to
see why it's important, right.

For ex I, I think that availability and
maintainability, I like to take an example

all the time as a bicycle example, right?

If I was asked to design a cycle
with a bike, with a requirement

which is the most flexible,

you can build like a monocycle, right?

That one with a one tire.

That one is pretty easy to maneuver.

You can just do 360 on it.

But it's not easy to ride
because balancing on it is hard.

And then there's tricycle,
which you can build.

Which has three wheels nobody's
going to fall off of it.

It's very easy to balance,
but it doesn't go as fast.

It's very hard to maneuver.

And bicycle is like a
balance with two wheels.

You can do balancing pretty well,
but you can also maneuver it.

But if you look at it, it's a spectrum.

Now you're making decision based
on what market you're operating

and what situation you're in.

If you're trying to really go fast
and you have a skilled guy who can

actually do with one wheel, that's,
that can only sustain for a little bit.

I think the bike is the most
prevalent architecture, right?

You look for flexibility and it's
a balance because you don't want,

also want a tricycle that doesn't go
as fast or can only cater to really

specific niche of, in this case, small
kids who are really trying to learn

to bike to get to the next level.

And so, yeah, it's always
a balance, I think.

Gordon Wong: I love that.

I love that.

I love that.

And that's part of, I think that's all
part of the purposeful architecture of

these products and these platforms is
that we have to understand the constraints

and we need to make investments, measured
investments in the right places, right?

You can't, you don't
get anything for free.

So

Sandeep Dhamale: That's

true.

Gordon Wong: What are
you excited about next?

What's coming?

What kind of emerging technologies
are coming that you can't

wait to get your hands on?

Sandeep Dhamale: I think I
kind of covered it with AI.

So I'll tell you one thing that I think
it's going to happen is especially in

the space that we're operating, all
of the data companies, and especially

companies who have data, like data
that either they generate or the data

that they have have been stewards of
for a long, long time, have a real

opportunity to think about what additional
intelligence layer that they can build.

So that's exciting in my mind.

And we operate in that space.

So very, very excited about that.

The second thing is a lot of these
models that are out there have

been trained on internet data
but not on enterprise data yet.

So I think that's going to give rise to a
lot of domain specific models and domain

specific use cases and healthcare is,

being in healthcare, I think that gives me
a lot of hope to solve for problems that

have been out of reach for a while now.

Or just because of time
resource constraints.

And when do you really get to it when you
have so many other things to solve, right?

So I think that's exciting for me.

And I think the third trend I'm
noticing is the private infrastructures.

Because a lot of people who have
these data assets are kind of thinking

about how can they not have to throw
this data out to these large language

models where they're not sure of the
security and whether their data is not

being trained for other purposes, etc.

I think that's where private large
language model infrastructures are gonna

play a key role in my mind and there are
vendors out there who are now supporting

with with these initiatives but yeah,
and that was also one of the things

that we did, just so you're aware, is
when when we build our infrastructure of

the large language model that I talked
to you about a couple of use cases.

With security first mindset we actually
because it was early days we got

the open source model LLAMA 3 back
then and LLAMA 2 and LLAMA 3 now.

And we have also looked at other
models to kind of bring on prem

we've spun up a GPU infrastructure
ourselves and have been training and

creating that infrastructure there.

The with the whole goal, and I think
with newer frameworks and newer

technologies, It's getting more easier
and easier to control your infrastructure

off these large language models.

It's still niche, but at least it's
possible that you are not now training

your data and model elsewhere, right?

You're, you can control
the, where that goes.

It can stay within your network
or within your preferred

partner network, if you will.

And I think that's exciting to me.

That kind of opens up a lot of use
cases because your security teams, if

they're not as worried, at least then
you can think about leveraging large

language infrastructure, large language
model infrastructure better for your use

cases because we have so many of them.

And I think that, that really was a,
you asked me about wins and brags.

I think that was one of the wins and
brags that I definitely am proud of that

we were able to tap into a private LLM so
early that kind of created possibilities

to think about some of the data use
cases or the content to data use cases

that we talked about and that's what
we'll be focusing on next year as well.

Gordon Wong: I'm excited about
the same things, I find myself

totally aligned with you.

Sandeep I've asked you a lot of
questions in this 45 minutes or so.

So I'm going to start a Newfire
podcast tradition right now.

I think the guests should get a
chance to ask the host a question.

Do you have any questions for me?

Sandeep Dhamale: Sure, Gordon.

First of all, it was
fun chatting with you.

I think we talked a lot about data
and AI and you bring a very different

perspective because you've been yourself
at the helm of so many different

initiatives across your career.

I think in your current role, you're
trying to help a lot of customers.

So some of the things that I talked to
you about are, are those the same things

that you're seeing across the industry?

I was just curious, like your
perspective, because you have a

vantage point like no other, and I'm
trying to get your perspective as

well on, on the trends that you're
seeing across both data and AI space.

Gordon Wong: Yeah, that's
a really good question.

So one of the things I think I realized
when I think about it is that what's

old is what's new again, right?

So often what I'm finding is that we're
building very similar, not that I think

it's exactly the same solution, but we're
solving the same problems now, at this

point in my career, as I was 30 years ago.

Because at the end of the day, what
we're trying to do is take data,

some kind of measurement, some kind
of, and turn it into information and

knowledge that allows better decisions,
better actions, and better outcomes.

Right?

And what we're seeing though is that as
the world's gotten more sophisticated

and we have more technology and we
have more capabilities, we're just

tackling more sophisticated problems.

30 years ago, it might've been trying
to measure who my best customer is.

But now we're trying to drive better
human health outcomes at scale, right?

We're solving problems that
we've never solved before.

And so that's what I think I'm
taking away from this and what I'm

reminding my clients at Newfire,
and we talked to them is that, that

yes, the technology is new, right?

But the problems are old in some ways.

It's just a different scale.

So since we know how to solve these
problems, let's bring that thinking, that

learning we've had in the past there.

You know, ROI, we talked
about ROI a second ago, right?

Numerator, denominator.

Okay.

So what are the costs that
we are incurring in trying

to drive better outcomes?

Okay, how do we burn those down, right?

What's the value we're driving?

Hey, what's holding back the value?

Where's the friction that is
keeping us from delivering?

If I tell my analysts, if you
deliver a presentation to your

audience you give them advice on what
they should do in this situation.

But they don't take that advice.

You haven't really
generated any value yet.

So part of your job as an analyst is
to be persuasive, is to communicate.

And part of persuasion and communicating
is also proving that our advice or our

data products are safe to use, right?

You used the metaphor earlier
about milk and groceries, right?

If I don't know this milk is safe to
drink, I'm not going to drink it, right?

And if our customers don't know that
or our customers' stakeholders don't

know that the data is safe to use,
they're not going to use it, right?

So I encourage our clients to
make sure they are investing in

that safety, that maintainability,
and make sure it's visible.

And so when someone asks,
is your product safe to use?

You say, absolutely, yes.

And here's how I can prove it to you.

Sandeep Dhamale: Yeah.

Gordon Wong: We're seeing these same
things across all our company, all

the companies and yes, a lot of use
cases in their details are new, a

lot of chatbot stuff, a lot of using,
models to predict outcomes, right?

And then we're making those outcomes
more deriving those predictions more

quickly, it's greater accuracy further
into the future, but also tying all

these assets together to get a more
holistic view , right, as opposed to

just these little point views, if I
use a simplistic metaphor, if I'm going

to the beach, I want to know how to
get there, what's the best clam shack,

what the weather's going to be, what's
the tide, I want it all in one place.

And

it.

now it's getting easier.

Sandeep Dhamale: Yeah, no.

Love it.

Love it.

And that totally resonates with me, right?

Like, these are the same problems
that have existed, but the reach of

our What the technology has enabled
is that it's now actually possible.

I always think in early 2000s, it was
really hard to build a data platform

that was, that can unify all of these
data assets with the advent of cloud

infrastructure that kind of makes your
life much more easier to think about

what it is going to look like, right?

Like you have technologies at your
disposal, which can hold variety of

data, which was harder back in 2000s
and with NoSQL and other technologies

that have matured over the last
few years, it's become possible.

Same story with Vector and AI.

I think it's another tool
in your arsenal that's

just gonna increase your reach.

Really excited with those possibilities.

And I totally love the images that
you use there, yeah, of the beach.

Yeah.

And let's with, with this, the beach,
beach hits home because I'm in Chicago

right now and it's pretty, pretty gloomy.

So I'm already thinking about
which beach I can be next.

Yeah.

Gordon Wong: IMe too, me too.

Sandeep, it was fantastic talking to you.

I really enjoyed myself.

I feel like I actually learned something.

I'm looking forward for us
to be working together again.

And I hope you enjoyed yourself as well.

Sandeep Dhamale: It was so much fun.

I have always enjoyed
interacting with Newfire team.

This is an amazing team.

Everyone I've interacted with this at
this team has been super smart and I loved

this conversation also about our data
infrastructure and how the AI initiatives

and how to think about data and AI
infrastructure in this fast moving world.

We should continue talking more.

It's always fun to chat with you.

Thank you for having me here.

Gordon Wong: Of course, it's a pleasure.

Thank you so much.

Have a good day.

Thanks for tuning in.

Stay tuned for more episodes where
we continue to explore the toughest

challenges and smartest solutions
in business and technology.

Like and subscribe.

Until next time, keep innovating
and solving the hard problems.

This is Hard Problems.

Smart Solutions.

The Newfire Podcast.