All right, let's make a start.

We've got a relatively large amount to cover.

So hi everybody.

Thanks for for joining the session today.

My name's Charles Ferguson.

I lead the product management team for Postgres at Microsoft

and I'm joined by Motion who will come on stage

shortly from NASDAQ to talk about how they've built their

board Vantage solution on a variety of Azure services.

So I was going to kick us off here.

It's a relatively short agenda.

Their motion's going to expand upon that in a little

bit more depth when he gets here.

But the basic content of the session today is going

to cover everything that you can fundamentally see on on

the screen here.

So NASDAQ are deploying these solutions using two of our

open source databases in in Azure today, Postgres and MySQL,

but then complementing that with a number of other key

services, most notably Foundry, where they're leveraging their AI models.

And Motion's going to walk you through, you know, how

they're using AI in the board Vantage solution for doing

a lot of work on documents and text to bring

a lot of efficiencies and optimization.

We'll talk about the models, talk about document intelligence and

how they implemented that over a previous way.

They were doing things to find a lot of efficiencies

and improvements in quality as well.

We'll talk a little bit about API management and then

Kubernetes, which is the the underlying application infrastructure for for

all of it.

Before we get into that, however, I just want to

give a few minutes here to talk about what we've

done specifically in Postgres at Microsoft over the last couple

of over the last couple of years.

Many people are often unaware actually that Microsoft is one

of the largest committers to the open source Postgres project.

And you know, Postgres has found itself in this position,

particularly over the last couple of years.

And I think you know, by virtue of the fact

that when the ChatGPT tidal wave broke, Postgres was one

of the only databases that have had a native vector

index directly within the database at that time.

And if you take a look at the popularity of

PG vector, it basically hockey sticked upwards from that point

in time, February about 2023, it was already had strong

developer preference and we've seen that continue to grow.

But amongst the hyperscalers, Microsoft has been the largest committer

over the last three years and it's part of really

how we approach working in the Postgres ecosystem.

We build and ship Postgres services.

We ship Azure Database for Postgres and we've announced Horizon

DB at Ignite as well.

But part of being a responsible member of that community

is taking what we learn from open source software, running

it at scale in the cloud and committing it back

upstream into the core open source project.

And that is exactly what we do.

It is a virtuous loop that just continues around.

It makes Postgres better for for everybody and it's how

we participate in that in that community.

Now at Ignite, we've had a couple of announcements specific

to flexible server.

I'm not going to drain the entire list here, but

two things that I really want to call out is

how we've lit up the latest generation of skews directly

into the flexible server service.

We've basically doubled the maximum number of E course that

are available for you.

We've made improvements to our storage as well, which are

not mentioned here as well, but that's enabling high availability

on a more performant version of SSD.

The other one that I just want to talk about

is in the lower left hand corner here, the post

Chris extension for VS Code.

So we shipped this actually, we announced this actually at

Pycon just before build and then re announced it at

build earlier this year.

I think we're close to about 280,000 installations right now.

We've seen a ton of positive feedback, a lot of

interaction with that community who have asked us to fix

little things.

We've been able to turn fixes incredibly quickly and we're

seeing really strong adoption.

And our vision with this extension in Postgres is to

offer effectively the full life cycle from provisioning to managing

to developing, leveraging Copilot for more productive, productive development and

then also debugging, monitoring and then migration as well.

And you've seen a lot more of that come to

light at Ignite this year in other sessions.

But if you do use VS Code and you're interested,

go take a look for the VS Code extension from,

from Microsoft.

It's got about, as I said, 270, two 180,000 installations

today and you should be able to to spot that

one.

Now the big announcement that we made Ignite this year

was around Horizon DB.

Horizon to be is our new cloud native Postgres service

where we've basically built a new storage capability that is

zonal resilient by default.

There was a great session given earlier this morning by

Denzel and Adam that went into a lot of depth

on how this works technically.

But basically, it's designed to deliver much higher transactional performance

throughput, much higher vector search performance with a full zonal

resiliency configuration so that you can run these mission critical

workloads with the highest levels of availability.

And then very simple failovers that sort of make it

a little bit more, improve the availability, reduce the downtime

relative to perhaps self managed Postgres as well.

So this is in private preview right now.

There is a sign up form that you can get

started on.

I think we've got a slide for that towards the

end of the session.

But really what I want to do is hand it

over to Mohsen.

So before he comes up here, we've got a short

video we're going to play and then I'm going to

give it to Mohsen for the rest of the sessions

to walk you through how they built board Vantage on

all of this Azure technology.

So thank you.

Where's my video and hit play?

This milestone marks the culmination of our future of the

boardroom floor, which is a celebration of bold ideas, transformative

technologies, and strategic partnerships that are shaping the future of

governance.

It's been such a thrill to be building AI driven

products with NASDAQ.

AI changes the way we work, it changes the way

we live, and now it's reimagining the way that we

can make the most important decisions at the highest level.

We are reimagining how boards operate, empowering boards to lead

with clarity and confidence in a world that demands agility,

insight, and precision.

Thank you very much Charles.

It is really fantastic to be here.

Hello everyone.

My name is Mohsen.

I am a senior manager of engineering at NASDAQ and

I work with the NASDAQ board Vantage platform.

I'm here today to talk about our AI driven governance

on Azure Foundry and postcard Sequel.

But before we go there, I'd like to walk you

through the agenda so you kind of know what you're

in for.

The picture Microsoft has used here is from a few

years ago.

A lot of things have changed since then.

One thing that has really changed is how we do

governance in in 2025.

That's exactly what we're going to be talking about.

We will talk about Nasdaq's cloud journey and some of

the challenges that we've had with that and how Azure

helped us to to solve those challenges.

We will then talk about enabling AI on our platforms

and some of the features that we've put into the

hands of our customers over the last two to three

years.

We will be doing a technical deep dive and I'll

be demoing a couple of features for you guys.

As a tech guy, I know sometimes in events like

these, the demo gods are not very kind to us.

So in case something goes wrong, Fred, not we have

backup videos, so you'll be switching to them.

But hopefully everything should be fine.

And then we'll be doing some architectural deep deep dive

to explain some of the architectural decisions we made.

And then eventually, if there's any learnings and evolution on

our platforms, we'll be talking about that as well.

So why does governance matter in 2025?

Governance is no longer about oversight.

It's actually about foresight because we are living in a

technical revolution that's being driven by AI.

The decisions need to be made really fast because stakes

are really, really high.

Governance is what is keeping leadership aligned, agile and accountable.

And now at NASDAQ we want to solve a problem.

That problem is where traditional ways of working are becoming

less and less relevant, where decisions have to be faster,

smarter, and be made with greater transparency.

This is where NASDAQ Board Vantage steps in and really

shines.

With our AI driven workflows like smart document summarization, AI

generated meeting minutes, contextual insights into your upcoming meetings.

We are transforming governance from a manual reactive process to

a proactive strategic advantage.

I'm happy to to say that since we've released our

AI features to production, our customers report more than 25%

of time savings annually.

That is more than 100 plus hours saved on manual

workflows like executive summaries of board documents and manually creating

meeting minutes.

So that's 100 plus hours saved.

It's it's, it's fantastic savings for for our customers.

However, it started with something slightly different.

It started with our move to the cloud.

Now I know this session is is mainly about our

AI foundry and postcard sequel in our AI workloads, but

I think it's really important to understand how we did

what we did and the role that our partnership with

Microsoft and particularly Microsoft Azure played in, in driving our

innovation forward.

Today, NASDAQ Board Vantage serves more than 4500 companies worldwide.

We are serving almost half of Fortune 500 organizations.

It is really important for us to speak the same

language that our customers do as most of our customers

are already in the cloud or are on their way

there.

As I'm sure is the case with all the organisations

represented in this room, you are in some way, shape

or form towards your journey in the cloud.

So when we are onboarding our customers, Azure really simplifies

that onboarding process because we can speak the same language

when it comes to cloud standards such as data privacy,

security, cross region deployments and everything that comes with the

with being on the cloud.

Now, NASDAQ is also a, like I mentioned, a global

platform, which means we operate in different parts of the

world.

And that while amazing, means that we have to comply

with regulations pertaining to data residency and data processing in

different parts of the world.

When we were in the data centre, we were managing

this by, you know, having data centres in different regions.

So being on Azure meant that that had to, to,

to meet that requirement as well.

And now you might be thinking, what is this guy

talking about?

Of course, being on the cloud means that we will

be meeting all these requirements, but let me explain.

NASDAQ Board Vantage today is made-up of many moving parts

from Azure Kubernetes Services to multiple databases to to BLOB

storages in Azure key walls.

And now with Azure AI Foundry where we are deploying

different models, all of these components have to follow that

exact same standard when it comes to a regional data

residency and and processing.

So for example, if I'm going to deploy a 4.1

mini model in Australia E, it has to be processing

it's data in Australia E for that customer.

And again I would I'm happy to say that Azure

has has really delivered in, in that aspect.

NASDAQ board Vantage is also a multi tenanted platform.

What that means is that for each of our 4600

customers or actually more than 4600 customers, we are segregating

them based on on databases and file storages.

And then on top of that, we are providing them

a extra layer of security by encrypting each customer's data

with their own key.

Now when we wanted to move to the cloud, it

had to to to satisfy these primary requirements of our

backbone architecture.

And I'm happy to say again that moving to Azure,

we found a one to one fit for pretty much

everything that we had.

So for example, when we were migrating our database, the

Azure database for my sequel server was a perfect fit

for our multi tenant system.

We we were we could migrate everything 1:00 to 1:00.

And then on top of that, we got high availability,

auto scaling, auto grow and we didn't have to worry

about managing our own infrastructure over there.

Then when it came to customer security keys, we were

using HSMS in the data centre.

Moving to the cloud, Azure offers AKV which is doing

exactly the same thing, but again, now we don't have

to worry about high availability and everything that comes with

maintaining the the hardware.

And then for for for file storages.

The Azure BLOB storage was a perfect fit as well

because now we could, in the BLOB store, store each

customer's data based on their own storage accounts.

But I would say the greatest benefit that we get

by being on the cloud is our engineers can focus

on writing and delivering business value without having to worry

about maintaining our own infrastructures.

I don't have to get into those late night calls

anymore to worry about oh we need to patch this

server or a hardware module fail.

Sometimes I still get called into those calls, but those

are usually because of the bugs I've introduced into the

system.

But I would say now with the Azure AI Foundry,

we have access to all the AI services as well,

which has really helped to drive our innovation forward and

help to stay relevant in this ever changing technological landscape.

I'm not going to bore you by explaining each part

of this document.

This is just to explain the gravity of of our

infrastructure.

You can see we have many, many moving parts and

components from Azure Kubernetes services to multiple databases, event driven

architecture.

Now AI, foundry, BLOB storage.

And obviously we operate in a high available, highly available

environment.

So each of these regions have a primary and a

secondary and all of these are replicated across all the

the the regions that we operate in.

So you can see it's quite an extensive architecture that

Microsoft Azure is helping us to to maintain, right.

Let's pivot.

As soon as we had moved to the cloud, I

think this was sometime in mid 2022.

Later that year we all know chat GPD came and

it changed the way that we are going to be

engineering permanently.

In my opinion now NASDAQ being the transformative company that

it is recognized the transformative potential of this technology and

we got to work.

But then again, we didn't want to just, you know,

go to market with the first feature that we could

have just for the sakes of being the first one.

So NASDAQ poured a lot of energy, time, resources and

a few dollars to analyse what would make the most

sense for our customers, our product people, our business.

They spoke to our customers, identified the the business flows

that would make the most sense.

And this is where we decided that smart document summarization

and the AI meeting minutes would be the the best

strategic advantage for our customers.

And they've been living in production for, for quite some

time.

And the next step that we wanted to focus on

was driving customer adoption.

I, I'm happy to say that, you know, out of

more than 4500 customers worldwide, more than 1000 have enabled

AI on their platforms already.

And you might be thinking that's less than 30%.

You guys need to do better.

But let me explain.

We have, we operate in a very sensitive segment of

the market.

Our customers trust us with some of their most strategic

sensitive data, data that is used to drive business decisions.

When AI was becoming mainstream, there was a lot of

challenges that we had to go through.

We not only had to solve those challenges, but we

also had to build that confidence and pass it down

to our customers.

So that's exactly where Azure and PostgreSQL has been helping

us.

But before I go there, I'd like to talk about

some of those challenges that we that we faced and

we have solved.

AI is non deterministic.

Now as engineers, we have worked with non determinism before.

We have dealt with large distributed systems where certain risk

conditions do not execute in the order that you expect

them to, but they are usually edge cases and they

can be mitigated with a significant amount of effort.

AI is inherently non deterministic.

That's just how it is.

We have to accept that, work with that and work

around that.

I don't need to impress upon anybody how bad hallucinations

can be, but in our line of work, we are

customers are again using data to drive decisions.

Hallucination can be catastrophic for both us and our our

customers.

So that's also something we have to understand and work

with.

Now, the medium that we use to communicate with the

AI today is natural language and prompt engineering.

And we know that natural language is ambiguous.

Again, as tech people, we work with programming languages.

Programming languages are built to remove ambiguity, unless it's JavaScript,

then it's just ambiguity.

But I digress.

The point I'm trying to make is that natural language

is inherently weak when it comes to removing ambiguity.

Also something we have to to understand and work with.

Now AI has come a long way since 2022, but

it is still slow and expensive to deploy at scale.

And organizations like NASDAQ with products like Board Vantage, which

are trying to mainstream that we have to be aware

of that and work with the with the limitations that

we have.

Obviously AI also introduced uses a new way of a

new challenge of testing and validating the outputs.

Again, as programmers, we are used to determinism in our

unit tests, but you can't do a assert equals on

the response from your, your AI output.

So we either have to build new tools or work

with tools that somebody else is, is building to validate

and test AI in this in this changing paradigm.

Well, I think that's enough for the theory.

Now I'd like to talk about some of the, the

AI features and show you some, some demos.

Now since 2022, we have put a few features in

the hands of our customers.

The ones that I'd like to talk about are meeting

minutes and the board assistant.

The meeting minutes is AAI feature that has we, we

deployed it in March earlier this year.

I do want to impress upon everyone that a board

meeting minute is very different from the meeting minutes that

you would see from a regular meeting.

These are legal doc records which can be used in

court proceedings.

These these contain important information about deliberations, how decisions were

made or anything and and basically strategic word exchange.

Corporate secretaries, after the meeting has happened, spend a lot

of time trying to formulate these meeting minutes to ascertain

their accuracy.

In fact, I was on a call with one of

the corporate secretaries a couple of weeks ago doing a

demo for this feature and she told me they spend

more than two weeks to to finalize these meeting minutes.

Our tool can do it in less than 5 minutes.

But before we go any further, I'd like to show

you guys how this tool actually works.

So I'll switch over to a demo.

So there's a few ingredients that goes into to building

these meeting minutes.

You have an agenda document.

I will upload that.

An agenda document, as the name indicates, will basically outline

whatever is going to happen in the meeting and who's

going to talk about that.

It's also going to mention any participants in the meeting,

whether they are joining online or they are joining in

person.

Then you have a meeting material.

This is basically a board packet, and these can be

massive documents, up to 400 to 600 pages.

Sometimes I've seen even more meeting notes.

Basically is the the the note taker in the room,

they will record any word exchanges that happen, voting decisions,

all of those things that are not obviously in the

board packet.

And then we allow you to upload an example of

the previous meeting minutes because we understand that each company

has its own way of writing, formulating their meeting minutes

document.

We don't want to assume that you want in a

certain way.

We let you upload an example and then we use

AI to apply that style and formatting into your your

linguistic formatting, basically into the actual document.

So I'm going to go ahead and attach an example

as well.

And then we'll press generate.

Obviously, we're not going to wait for this because it

can take a while and we don't have a while.

So I have previously generated a document that I'd like

to show you guys.

And you can see that this is a meeting minutes

from a Massachusetts for authority.

Priority meeting and I'm not going to go through the

whole thing, but you can see how the AI formulated

the the entire document.

Let me just make it a bit bigger.

And it will based on the agenda that you've uploaded

is going to Add all the information.

And then in the end, if there's any key decisions,

it will add those as well.

And then basically it's formatted according to the, the example

of the meeting minutes that was provided, right.

So let's talk about what happened here.

I uploaded a few documents, some magic happened and then

you had a draft of the the meeting minutes generated.

Let's try to unravel that magic as I uploaded the

agenda.

We use document intelligence and I'm going to talk about

document intelligence and the importance of that in just a

second.

We extracted all the information from from the agenda and

then we applied the agenda processing prompt.

Obviously, this is a special prompt that parses the agenda

in a in a way that we're going to use

it further.

We process that with 4.1 mini and then we store

that in the in the memory for for the processing.

As a next step, we then extract meeting materials.

And this is where document is just becomes really important

because as I mentioned, these can be documents of up

to 400 to 600 pages.

When we did not have document intelligence, we wrote our

own semantic chunker.

The idea was that you want to create small chunks

of these documents, create embeddings and then store them into

the into the database.

That chunker was doing quite well, but it was always

a guessing game where a section starts, where a section

ends and sometimes it would not be very accurate.

When we started using document intelligence, basically what it does

is it gives you a well defined layout of the

entire document.

So you know where it's the heading, you know where

this a section starts, a section ends.

And it's particularly very good when it comes to recognizing

figures and, and other diagrams because sometimes a lot of

context is missed in those things.

So once we've done that, we still perform the chunking,

but now it's easier to to chunk this this document.

And then we create embeddings with the text embedding ADA

and persist that into the the Postgres sequel.

We will be talking a lot more about the Postgres

sequel with PG vector enabled in just one second.

The exact same thing happens with the with the meeting

notes.

We process that with document intelligence, semantically chunk it, create

embeddings stored into the Postgres SQL database.

And then obviously if you have an example of the

document, we will process that with the document intelligence and

we will apply a style guide prompt to it and

we will keep that on the on the side as

well.

Now that we have processed all the ingredients of this

this massive board context, it's time to start putting it

together.

So how do we put it together?

We take this processed agenda and then we match each

of the headings to the stored embeddings.

And you can see that with Postgre SQL with PG

vector enabled, it is so easy to run a hybrid

search.

What you can see down here is a Java code

snippet with a SQL query that has a vector similarity

search as well.

You don't need to.

There's no rocket science here.

It's just simple SQL with hybrid search.

And then you can do a look up based on

the cosine similarity.

And then you can limit up to the number of

embeddings that you are looking for.

The exact same thing happens now that we've matched the

embeddings, we can apply the prompt because remember, what we've

selected right now is basically chunks of text, right?

So when you apply the, now we need to process

it with natural language so it makes sense to the,

the reader.

So we apply the prompt, process that with the chat

completions.

And now you have the agenda that has been enriched

with the meeting material.

And now the exact same thing happens with the with

the meeting notes, same process, refine it and then add

that to the enriched draft that you had with the

with the meeting material.

And now if you have a style example optional, but

if you wanted, we are going to apply that style,

apply the prompt process that which had completions and you

have a final drafted meeting minutes.

I must impress upon each of you that this tool

is not built to replace the corporate secretary.

It's built to make their lives easier.

So instead of spending two weeks on drafting this manually,

now they spend less than 5 minutes and they can,

you know, be more productive with their time.

And then obviously you want to have the human in

the loop.

That's why you get a Word document downloaded.

The corporate secretary can go back and edit that if

she wants, or he wants to add more information to

it.

Or if anything doesn't make sense, they could change that

as well.

Now with AI meeting minutes, we proved that we could

use PG vector enabled in in Postgres Azure database which

is natively available.

You all can go and do that right now.

We prove that it works for production workloads and it

was working really well for us, for all our customers.

So then the next natural step was OK, how do

we use it in our next feature, which is the

the board assistant.

Now the board assistant is infinitely more complicated than than

the meeting minutes and it will be fun demoing it

to to you guys.

But the idea with the board assistant is that we

want to make life easier for for our users, particularly

our directors.

So if you are a, if you're a director heading

to a meeting, you're probably busy with a lot of

things.

What we want to do is provide you contextual insights

into your upcoming meetings.

So for example, it'll tell you what you should be

preparing for based on whatever has happened in the last

two to three meetings.

And then it gives you a chat assistant that you

can use to, to talk to and get more information.

But let me show you how that works.

I know I have to switch the screen.

Just give me one second, all right.

So imagine that I am a director.

I come into my board portal, I have a meeting

on the 20th of November and then I can see

an AI icon over here.

I click on this guy and then it tells me,

it shows me the different insights that will be that

I should be preparing for my upcoming meeting.

So for example, strategy, risk, finance.

So let's go ahead and click on risk.

It has selected the top 6 insights for Risk that

I should be focusing on in my next upcoming meeting.

So let's say I go ahead and choose cybersecurity because

why not in Signite.

So what it's going to do is it's going to

give me a comparative analysis of what we have talked

about in the previous minutes, previous meetings and what should

we be focusing on, on the upcoming meeting, right?

So for example, I'm not really happy with what it

has told me.

I could say, I could start talking to it and

say, could you expand on Queue 3?

And just like that, it's going to go and look

up all the information stored in the in the back

process that with AI and give me what I should

be focusing on.

You can also see that it gives you follow up

prompts.

So based on the conversation you're having, it will tell

you what other questions you should be asking.

So you can ask that question as well.

Then on top of that, if you want to know

something else about the meeting, for example, let's say who

else is attending, could you tell me who else is

attending?

So it's going to give me information about who else

is attending.

So if I want to avoid somebody, I will not

go to this meeting.

And then obviously you can switch to to the other

insights and you can do the exact same process over

there as well.

So let's switch back and see what has happened.

Now, as I said, our board assistant is much more

complicated.

What's happening here is we are just introducing building this,

this new feature, but we have had customers who've been

adding data to our platform for the last 8 to

10 years.

And this is a lot of data, right?

So if you are, if you have a, a board

packet, as I mentioned, this could be like 400 to

600 pages.

And then multiply that by the amount of board meetings

that you're having.

It's a ton of data that needs to be processed.

So what we did is we built a, the board

assistant in two parts.

1 is your data processing pipeline.

So as a customer enables AI, you're going to go

and we we look up this data process that with

a pipeline.

We'll just get to that in a second and then

store that into the the vector database and also the

my sequel for for non non vector information.

And then the board assistant part is basically the interface

that lets you interact with the assistant.

As you just saw in the demo for the data

processing pipeline, we decided to go with a serverless approach.

The reason for that is because as I said, we

are processing an immense amount of data.

I don't want to worry about CPU, memory limits or

anything like that.

I will get what I need to, to, to, to,

to get to process this.

And I only pay for what I have to pay

for.

And if I'm not using it, I'm not going to

pay for it.

And then I don't want to, to put stress on

our existing infrastructure by processing all this ton of data

with the existing micro services.

It just it's, it's not meant for that kind of

workloads.

And then So what we're doing is we are using

a fan in, fan out approach.

There's an orchestrator that fires up per client.

I'm not going to bore you with all the activities

that happen over here.

But then for each of the meetings, it processes each

of the the, the the steps that are in part

of your meeting as activity functions.

Let's take a look at what happens in one activity

function.

So what happens is, let's say I'm processing the metadata,

so any information about the title, where the meeting is

happening, any other tags that might be important, I take

them, I chunk them, I create the embeddings, we tag

the embeddings so that we can find them later in

a in a easier way.

And then we persist them into the the vectorized database.

The insights generation that you saw is a preprocessed activity.

We have a research agent that runs in the background

that's going to pick up all your documents.

It's going to run a comparative analysis, it's going to

do a, it's going to refine that and it's it's

going to run a comparative analysis again.

And it does that over many passes until it's reached

the state where now you can say, OK, I have

a very good comparative analysis of how a trend in

my insights has changed over the last three to four

meetings.

And then obviously this is not a vector.

It's basically text that will persist that into the My

Sequel database.

So you guys saw me interacting with the with the

assistant.

What happens is I write a prompt, it gets converted

into an embedding, we match that with the existing context

and then basically it fetches all the existing data.

The the reason why I put this piece of snippet

over here is thisis.net Core.

This is again to demonstrate how easy it is to

write a hybrid search query in any platform basically.

So all this is doing is employing EF Core link

and simple vector search and now you can basically limit

this with cosine similarity in any of the semantic rankings

that you want to do as well.

So once we found the the matching embeddings, we will

add the historical context so the AI knows the flow

of information and the context that the chat was being

held in.

We will then apply the prompt because obviously this is

a board assistant.

We don't want you asking about the weather or anything

like that.

We will then process that with a chat completions model

and then we are using the Azure service for Signal

R to basically stream this back to the back to

the client.

This is another view of how the the the chat

assistant works.

You write a query, the orchestrator kicks in, is going

to go look up all the information that it needs

from the vector database, combine that with the historical context

and then pass it to the LLM to process it.

And then the response is streamed back to the the

user.

If you look at the architecture, it does look quite

daunting, but it's actually basic Azure services.

We're using Azure Kubernetes Services.

This is where we pull all the information from the

into the into the functions app and then the board

assistant micro service basically is the one that is doing

all the the chat work.

Then we're using Azure Messaging which is basically we're using

Service bus to drive our event driven architecture.

Signal R is used to stream data back to the

client.

We're using Postgres SQL and my SQL to persist data

as you just saw.

And then we are using durable functions so that we

can keep track of what is happening.

And then if an activity fails, we can re trigger

that activity.

And then obviously Azure, the AI foundry is the backbone

of our entire AI infrastructure.

We are using multiple models.

We're using the 4.1 for chat completions, we're using the

O3 Mini for reasoning and then we're using the the

text ADA embedding 002 model.

And then obviously document intelligence, as I explained earlier, is

used heavily to process all these these documents.

In the beginning, when I was talking about enabling, AII

mentioned some of the challenges that we are facing, especially

related to determinism and hallucinations and all of those things.

I do want to talk about how Azure AI Foundry

has helped us to solve those.

If I don't know if you've used the prompt flow,

but I've found it extremely helpful to breakdown complex problems

into smaller steps.

You can write a prompt, you can add context, you

can write more prompt, you can add more context.

So it it helps you tackle determinism.

Now I am by no means saying that it's going

to remove determinism, non determinism, that's probably never going to

happen, but it helps you tackle it and helps you

engineer your, your, your, your systems in a way that

can be a bit more predictable.

Document intelligence, as I explained it has really changed the

way that we provide context to the AI.

Our context is much richer.

It makes gives us much more accurate responses.

Model deployment helps us to deploy any models that we

find relevant into our data zone deployments.

As I mentioned earlier, our requirement is that we will

deploy the model wherever our customer is based.

And then obviously we can configure our consumption plans as

well.

Now, as prompt flow has helped us in to tackle

complex problems.

Chat playground helps us tackle less complex problems.

But if you're looking for, Oh my God, what's the

output going to be?

If I, if I write a small prompt and this

is my context, you can easily get the response there

without writing a whole program that communicates with, with, with

open AI using the, using the API.

So it makes your life really, really easy.

And I use that actually all the time.

Then with Postgres and PG vector, as I mentioned earlier,

this was really a game changer for us when it

came to vector databases because we are using Postgres today

already.

So and Azure supports natively enabling PG vector.

So we enabled that.

It really drove our innovation because we could quickly start

building PO CS and then start building production workloads as

well.

Since it's already part of our infrastructure.

We have it in high availability, cross region replication and

everything that we have today.

We don't have to worry about the, the, the maintainability

and all of those things that come with again, maintaining

your infrastructure.

PG vector again game changer.

As I mentioned with with meeting minutes, we were, we

are storing data in a temporary way.

So it's, it's all once the the the flow is

completed, we delete all the embeddings with chat.

So I mean, in in this case, we didn't have

to worry about indexing because it's, you know, you, you

save it once you look it up and you delete

it with board assistant, it is much more complicated because

now you're adding millions of embeddings per customer.

So the H and SW indexing really comes in handy

over here.

And then combined with the the sequel search that we

do it, it really makes the queries perform much, much

better.

I think you would be thinking, wow, this is cool,

but show me the code.

Unfortunately, I won't be able to show you the code,

but I will be able to show you the next

best thing.

I'd like to show you a slide from or a

dashboard from how our AI systems perform in production today.

So the and and how these help us to basically

build confidence in our customers.

What we are doing is we are using deep eval

to evaluate our AI responses over many, many publicly available

government documents.

So for summarization flows and AI meeting minutes, we take

publicly available documents, we churn them through our systems and

we, we, we have thousands of different data samples that

we then process.

And I'd like to show you guys in terms of

different metrics for coherence correctness, you can see how well

these these systems are performing.

We check for bias, we check for toxicity and then

for answer relevancy.

You can run it per dock document.

We can do document cosine similarity score comparison and then

obviously you can do a a simple cosine similarity score

comparison over multiple models.

So we can switch the model, we can run the

whole thing again and it it basically will give you

the will give you the results and then you can

compare them to other models that that are performing right.

I mean, I would like to take this opportunity to

at the end, thank Microsoft very much for inviting us

here and having us share our experiences with you.

I think this partnership that we've had has really helped

to drive our AI innovation forward.

I've been working with Charles and Bart over here over

the last one year and they have been extremely helpful.

So I mean, this was a great experience for me

to come here and show you guys how our tools

we've built our AI tools.

And if you have any questions, I'll obviously be be

outside in the in the expert area.

Would love to hear from you guys.

And I would say thank you very much for your

attention and hope to see you guys again.