Introduction

Hi, everybody. Thank you, Chetan. He invited me yesterday, literally at the same time to come and speak here.

He's a great person. I've met him. I've actually never met him in person before. This is the first time I'm meeting him, but we have been chatting on WhatsApp a few times before.

Sheikh, when you were presenting, I thought he's taking away my whole presentation because I'm more on the technical side, but he's on the business side but he was talking about how numbers matter and I basically helped get to create those numbers create those dashboards that's what I do I'll talk about yeah thank you everybody coming here

Why AI is now embedded in the data world

on a summer very hot day so my presentation is about how AI is actually embedding in the data world everybody used to talk about that two three years ago how disruptive it's going to be how how complex it is, etc. Somebody asked a question on the governance style as well, that how you put guardrails, etc. Now AI is actually showing up in data world as well. So I'm going to talk about how I use AI in data world on a day -to -day basis.

Who I am and what I build

So who am I? I am Nitin. I basically build data platform from scratch for small to mid -sized companies.

I've done that for many years now for 15 years here it is this is my company i started my own consulting company after working in many companies so far so this is my company seeded insights .com focused on

specifically what it said this is my open source project anybody interested please check it out it's basically how in data world somebody was talking about tokens how tokens can start up to to add a cost how data platform tools you have can be costly that's why i use the name costly where

all the features you have all the dashboards you're building all the ai tokens you're using you should know exactly how much you're spending and this will even tell you how to optimize and save cost as well it started as was one of the features was on snowflake to reduce the cost on

AI across the modern data stack

snowflake what this presentation is how in last two years things have changed in the last two years overall architecture is same even think about governance etc everything is exactly same pieces are always all there exactly in the same location but it is changing

Ingestion: AI-assisted coding and debugging

in this today's world is this way where in each layer ai is actually integrating where if you look at ingestions ai is actually integrating and ingestion layers as well where it is you're helping you write code do help you debug so it's helping you do that ingestion part

Warehouse and modeling: fundamentals still matter

warehouse is still warehouse you do need to have a warehouse you cannot get rid of it

model there is your data modeling if i know there are five people who said they understand data world more or less modeling is where you're actually creating your facts dimensions so it helps you create those but it is not a business person can define what those facts and dimensions

Dashboards: more demand, more natural interaction

are dashboards are still here they have not gone away I think now because data is so much available that business actually wants more dashboards it's not less it's actually more but they're able to interact with the data in a more natural natural language processes so it is sort of nice it's not taking jobs away I think so it's adding more jobs how when I'm developing any platforms

platforms nowadays how new tools like tableau omni hex how are they really adapting to ai oh sorry this is a different one so this one is a very good example i want to talk about this

Case study: migrating legacy dashboards faster with AI

specifically enterprises have legacy now legacy is business logic built in they have 600 dashboards built out how do i really migrate migration is one of the projects which consulting companies will take on take years to finish the time they're finishing in there is a new migration happening.

This is one of the problems every company go to have faced more or less.

I was working for a project where they had 600 dashboards to be migrated into a new platform from Tableau to Omni. I can guarantee that would have taken a year to migrate, but we were able

to actually finish it in one month overall.

The business logic which existed in the GitHub Git labs, we were able to translate into much more natural language within a week or so time. I think that was a beauty of AI here.

The amount of time taken was way faster as compared to how it was before AI.

How users interact with data today: natural language to SQL

That's the beauty of AI. And this is how users are interacting with data nowadays on a regular basis.

They're asking questions in natural languages. It is converting into SQL query. SQL is still there.

If you think you don't need to learn SQL, I don't think so. That's true you still need to know sql to comprehend what the sql has been written otherwise your dashboards are not going to be correct they're going to be wrong because natural languages are converting your natural language into sql and actually creating a dashboard for you this is how business leaders decision makers are looking at the dashboards nowadays

The tech stack is expanding, not replacing what you have

nowadays the tech stack is changing but it's actually adding more layer to it so i would say we have our AI brain added on, but previous legacy tech stack still remains as is. DBT is there. Airflow is there. Nothing is gone. There are more tech stack added on.

Tool spotlight: a quick look at Hex

I want to talk about one thing real quick. Hex. I wanted to show you guys a very small overall demo.

What Hex is and why it stands out

This, I don't know, anybody heard of Hex so far ever? Tableau? Tableau, yes. Hex,

Hex? No. Hex is nowadays more of a leader in this space where they're coming out.

You can interact in a natural language and get a dashboard created. Earlier, Tableau, Looker, they have it. I won't say

they don't have it, but they're not so advanced how Hex is at the moment. So you can see the interface. It's pretty easy.

As soon as you look at their dashboard, sorry, at their landing page, you know what you have to do and you can get your work started. I'm not going to write it here year because of what I pre -prepared something for you guys where I was able to ask Hex a very

Speed vs. accuracy: why humans still need to validate

relevant question which on a day -to -day basis you probably will take a lot of time to figure out what your data is under the hood for you to figure out write a query comprehend it it takes a lot of time he talked about finding that number it takes a lot of time to do that for a human AI is helping accelerate all of that but is it accurate or not accurate that's where the human comes into the

play to make sure that it is accurate or not accurate i've seen a lot of hallucination happening and it can be a disaster for a company reporting on a wrong number you want to show so this was all written by ai i did not write any query it was all written by ai for me where it

Clean data and single source of truth as the foundation

was able to generate something and it generates some of the dashboards for me as well i did not do anything this prompting all right it is obvious if you put in clean data that's the core butter of data platform if you put in clean data you will get your answers right if you do not put in you will get your own numbers wrong and that's in i emphasize that a lot when i'm working on a

data platform that's one of the core element to it and i want to talk about very specifically

Three questions to diagnose trust and value in your numbers

when i go to any business meeting any business leader what type of questions i'm asking them to understand where the problems really lies he talked about numbers i think this is what i ask them every time do you trust your numbers if you do not trust it i know where the core problems are your data platform is not set up to have the right audit mechanisms or single source of truth

anybody technical enough to understand that there is a single source of truth concept if you do not have that you will not have your numbers right it will always be a problem and ai is going to to be hallucinate based on those numbers so repetitive task I think in AI sorry data world

there is too many things you have to do one of one of the examples I want to give is a lot of people operate on excel sheets csv files etc you always try to think of I can upload it it will get inserted into one of the databases etc it can be automated via AI nowadays it was a very ad hoc very engineer focused task nowadays I don't think so it is an engineer focus

so you think of those repetitive tasks which are doing very manually you can automate a lot of it so this is type of questions which business leaders are basically asking for what type what is the answer you're looking for from the

data you're trying to interact with so these are the three questions I start with that's where help me narrow down what the real problem is in the data platform.

Closing and questions

So this is me, open to questions. I'm based out of Toronto.

By the way, I did not mention that part. I live in Mississauga, not in Toronto.

Thank you.