Discussing Data Trends in the AI Era

Developers need a toolkit to build AI into their applications – models, functions, & data, Hypermode does just that. To learn how the world of data and databases have been rapidly evolving for the age of AI, we sat down with their CTO Gajanan Chinchwadkar (previously Chief Systems Architect at Visa).

Join as we discuss:

• The evolution of database technologies from niche markets to a $100 billion industry with over 8000 options.

• Scaling databases in different organizational contexts, and strategies for staying up-to-date with industry trends.

• How AI will impact the programming landscape.

David Joy:

What is up everyone? And welcome to the Big Ideas in App Architecture podcast. In today’s episode, I speak to Gajanan Chinchwadkar, who is the CTO at Hypermode, a company that is focused on building AI and helping AI based experiences for developers. In this episode, we get into Gajanan’s experience of working with multiple databases across the last three decades and get into his perspective on how things have changed and what is still essential for running databases in the industry. So pump up the volume and get ready for an intriguing conversation with Gajanan Chinchwadkar.

Gajanan, welcome to the podcast. How are you doing today?

Gajanan Chinchwadkar:

Yeah, I’m doing great.

David Joy:

You are right now based out of California, that’s where you are right now?

Gajanan Chinchwadkar:

Yeah, I’m in California in the San Francisco Bay Area.

David Joy:

Gajanan is currently the CTO at Hypermode. And before I butcher your introduction, Gajanan, why don’t you tell the people a little bit about yourself, and your current role at Hypermode. But before you get into Hypermode, tell the people a little bit about you.

Gajanan Chinchwadkar:

I’m Gajanan Chinchwadkar. So I have been in this San Francisco Bay Area for the last 24 years now and about working in databases for a long time. And interestingly, I have stints at many kinds of databases. So the way when I look at my past, I think that it was a very interesting journey from object-oriented databases, relational databases, SQL, XML databases, then pure XML databases, and tech search, and a combination of navigational queries, tech search, and temporal queries. And interestingly, I have been working mostly in database query engines. So even in databases, I was fortunate enough to get variety of experience and a variety of data models with respect to query processing,

David Joy:

I wanted you to expand on your current role as well as Hypermode, a little bit about what you do with that.

Gajanan Chinchwadkar:

I’m a CTO at Hypermode. Hypermode is AI space, a company in AI space, and it basically helps our developers to bring AI into their applications. So we provide toolkits for models, functions, and data together. That way, AI becomes accessible to the application developers.

David Joy:

So when I was going through your profile, obviously you’ve had a seasoned career. When I was just going through LinkedIn, you have worked at so many amazing companies and across multiple databases, and that’s what we hit off on when we connected. But I also saw that you started your career as a professor back in India in a place called Sangli, which I know about because I did my engineering back in India as well. So how did a shift from academic to industry happen for you?

Gajanan Chinchwadkar:

So when I was teaching, it was undergraduate level teaching, and I had a master’s degree, so I applied for a PhD. But during my teaching career, I realized that it is very important to understand the real life problems in the industry. And so towards the end of my PhD, although I had offers for academic jobs, I decided that let me at least work for next few years, make five years or so in the industry. But once I entered the industry, there was no going back. There was no way to go back. I was enjoying my work so much that I never thought about going back.

David Joy:

Tell us a little bit about how you got into the database side of things. What inspired you and you felt like, well, that’s the stuff I want to work on?

Gajanan Chinchwadkar:

This is very interesting. Some of that is coincidence, and some of that is inclination also. So when I was doing my master’s degree in India, I was working, that was a collaboration project between IIT and Center for Development of Advanced Computing, CDAC, which was a prime institute, a very reputed institute in India building parallel computers, first parallel computers. So I was working in mapping of parallel programs on the parallel architecture. So it was purely a graph theoretic kind of problems, because parallel programs can be represented as graphs and the links can be represented as communication between different modules, and even the architecture is a graph, and there is a network between the CPUs. So this actually attracted me to some of these graph theoretic problems. And so parallel and graph [inaudible 00:05:13], that kind of interest. When I entered my PhD, I was inclined that way.

I went to NTU in Singapore, and my supervisor and some of her students had interest in object-oriented databases. When you look at object-oriented databases, particularly the data model, the query processing problems, all of them looked like graph problems, and that’s how I started working in object databases. Eventually, because I then graduated with a PhD in object databases, particularly in database field, I was looking for a job in database field, and my first job I landed in Sybase. They had started a new development center in Singapore for developing query engine related work for ASE. Sybase ASE is one of the famous products basically, of Sybase, Adaptive Server Enterprise. And there I got into nested sub queries and then view materialization, view processing, view flattening, all those kinds of things, and they were so… I started enjoying that so much.

These problems were not only complex, but they were also a little bit academic. So it’s not just the programming or coding, it’s a lot of analytical skill and understanding various age cases use the various possibilities of bugs. And especially you can imagine some hard bugs, I was actually, I used to get thrilled, once I came across a bug which was a 30 level deep nested query, and it was returning wrong results. So it is just a pleasure just to simplify that problem and trying to understand at what level in this query the problem is happening and why it is happening and how to fix it, so that you don’t break anything else.

David Joy:

Yeah, you make a really good point. That world at that time in the industry was such a thrilling time to be in. I know there was a .NET era, a .NET boom happened, and I think there was a dip, and then everybody started building applications of personal computers growing, the internet was growing. Google search was coming up at the same time. Of course, I came after all of that, but I’ve followed, as somebody who’s interested in tech, to go back and learn the history. The few databases that existed, Oracle was leading the chart really, and you could count the number of databases that were in the market. But today, I don’t know if you follow, a list came out, we have almost 8,000 databases in the market and it’s $100 billion cap overall market now, so it’s such an evolution.

Gajanan Chinchwadkar:

Maybe I was fortunate to work on real enterprise databases, and there always used to be few of them, and there used to be a lot of rigor and many customers using them. As a result, you would see many variety of problems. And while designing also anything new, you have to consider so many common cases and use cases and how, because tools used to generate SQL queries, so you have to think how certain tool can generate something, those kinds of things. Today what has happened is that there are too many small databases. They might be getting used in some pockets, but most of them are really not that prominent still. I think there are only, you can still count them basically on the tips of your finger that, among the newer ones, including CockroachDB is one of the popular ones, or Couch, or DataStax, Cassandra. There are just a few actually. MySQL, Postgres, among the open-source ones. Yeah, they’re just limited ones. Most of them are, it’s like I sometimes think it is a nice, just too much of nice.

David Joy:

Tell me a little bit about, let’s dive into a little bit more about Hypermode and why Hypermode, like in the sense… I know you got into Hypermode through an acquisition, if I’m not wrong. Tell me a little bit about what you felt was the problem and why you started Dgraph, which was eventually acquired by Hypermode, from what I understand. Tell me a little bit of background on that.

Gajanan Chinchwadkar:

So Dgraph was actually a graph database company. This was founded in 2016 by an ex-Googler, basically. And so at some point the investors of this company approached me, asked me that, can you take the technology leadership of this company? And then after some back and forth, I decided to join the company because, again, my fascination towards graph from my college days. So I joined this company, but at that time, because this was very recent, I joined Dgraph in 2022, I could clearly see that the new AI wave is coming. And also, I could see that in this AI wave, databases need to be different. I won’t say that relational databases are not important, I would say that they are probably the most important still, and even in the AI wave, they will remain the most important databases, but something else is needed in addition to that. And what is that?

So the wave of these vector databases recently proved that, yeah, there is something needed for semantic searches. And because I had worked on keyword searches and full-text searches, I knew that there are limitations, in the past also. But I was still thinking that this whole thing in AI, all problems, data, data about data, data about models, all that is really represented well in graphs. And so graph databases will play critical role. That was the hunch on which I joined this company.

But then once you join the company, you realize that, and you being a salesperson, you know that market sales of databases do not happen based on the technology only. There are a lot of factors which go into consideration for purchasing databases by any organization. So naturally, the graph database market so far is still small, but the place of graphs, and particularly a combination of graphs and vector, which, in mid 2023, I started thinking about that when we have already a powerful graph and we just add a vector to this, this will become a perfect kind of a database of storage for many AI applications. I won’t say every AI application again, because there are so many applications which are written with just based on MySQL or Postgres relational databases. That is the fascinating part actually. I still get excited to see the possibility of graphs and vectors, semantic search, all that in this whole RAG pipeline, and how it’ll enrich the whole GenAI ecosystem.

David Joy:

You brought a very interesting point because I have been myself following some of these graph databases just as an enthusiast of an AI person. So I’ve been following Quadrant, I’ve been learning about what Pinecone’s doing. Obviously with CockroachDB, we do hear about pgvector and pgvector support. How is it that what you do at Hypermode different from all these existing databases that also do vector embeddings and things like that as a feature?

Gajanan Chinchwadkar:

Oh, okay. So let me tell you one thing. So then this company, this Dgraph got acquired by Hypermode. So then what is Hypermode again? We come back to that question. Because Hypermode is not a database company, really. Hypermode is a company that enables application developers to bring AI into their applications. So what we provide is models then functions, some built-in functions that they can use in their applications, ability to add custom functions, ability to connect to data sources, or ability to ingest data. Similarly, ability to connect to the models of their choice, all that stuff. So in this whole game, there is a lot of metadata, and also there are many RAG-based applications which need vector support. So the Hypermode is not going to dictate what database people should use, we’ll not necessarily will say that they have to use Dgraph, but internally we’ll be using Dgraph because it could power a lot of our applications. And also, some of the technology from Dgraph will help in AI applications.

David Joy:

So I wanted to jump back to one of the things that you were talking about, is that you being a veteran in the database space is that databases have pretty much existed to do the same function, store data. But now the form of data has changed, you store data in tables, store data in strings, or store data as embeddings. At the end of the day, we are storing data. But you made a very good point that traditional databases, or OLTP transaction level conversations are never going to change, or will mostly remain the same. Why do you feel so strongly about that?

Gajanan Chinchwadkar:

Because I have seen this journey for a long time. When I started early initially at Versant, which was probably the topmost object-oriented database, in those days, there was literally a standards committee which was working on object theory language standard, and they came out with two or three versions of that standard. So OQL, that was called OQL. It was never accepted by people.

Then came the whole era of XML and XQuery and XML Schema and all that. It was not accepted. So what is going on? Fundamentally, it is easy to learn SQL, and there is a large pool of developers who understand SQL. And not only developers, even sometimes business users also can write simple SQL queries and get their tasks done. So that is not going to change, basically. Even if you look at the recent developments like these data frames, or even Databricks, Spark. Spark is kind of supporting SQL, isn’t it? And the reason is that I don’t think originally Hadoop or Spark, they were intended to be SQL engines, isn’t it? But Hadoop then added Hive, Spark then added Spark SQL. All these things happened because of that option. And not only that, if you look at these NoSQL databases, you see the way they initially, they were saying literally NoSQL towards. Okay? Towards the intent, about a decade, they started saying not only SQL.

David Joy:

That’s probably somebody sitting in the marketing department.

Gajanan Chinchwadkar:

So naturally there is a reason to believe basically, it is impossible to change so much of investment that the industry has done in relational databases. It is nearly impossible to get skills which are other than SQL language in the market for companies to build their applications. With all that, there is reason to strongly believe that relational databases are going to stay put for a long time.

David Joy:

No, I think you make a really good point. Because this was also, I worked at DataStax before, so I know the Cassandra world obviously, and I felt like these NoSQL databases, especially that came out of those papers that were written in 2006 and 7, fit and brought into the industry and came at a time when we required scale, data was exploding, and we needed to scale horizontally, and they fit that infrastructure requirement of scale. But we had to have a trade-off, the trade-off was consistency, and the trade-off using denormalized tables and building your application to get exactly what you want.

But then I started doing this when I started working on some projects on the side, I wanted to do some basic things with SQL, like just do this or that with the SQL paradigm. Because even today in colleges, you’re still teaching SQL or basic SQL to everybody, and that’s what every student learns when he comes into the industry. The only exception to this is probably MongoDB, because they have built a developer-focused database, obviously, but still doesn’t solve the use cases that require OLTP transaction and scale, especially for tier zero, tier one kind of applications that high performing mission-critical use cases require. So I agree with you, at the end of the day, the industry knows that SQL is the standard and we have to support SQL, or SQL. So I appreciate you bringing that point of… Yeah.

Gajanan Chinchwadkar:

Yeah. And see, that’s the precise reason why I say why what excites me at Hypermode. People are not interested in learning a database and solving a problem using a database, people are interested in solving a business problem by building an application. So all these things have to be abstracted out. What database you’re using internally, it really doesn’t matter. But the interesting part is that even if you are looking at say, abstract functions, but they’re connected, they’re like graphs, and then you have to wire them, build pipelines, which is very similar to the ideas implemented in relational databases. Okay, so engineering wise, those ideas are similar, so technology or engineering ideas can be reused, but it is not necessary to be tied up with Dgraph or any particular database.

David Joy:

And I agree with your point on abstraction because I think that change we are seeing across the industry, even in personas, we have architects, we have developers, we have database administrators, three different functions at least. I don’t know if you follow this, there’s this company called Cognitive.AI that recently released the first software engineer, basically they call it AI software, and it’s called Devin. And basically what that AI program does is it acts as a junior software developer and it can basically do 90% of the job that even a senior software engineer can do. But the whole idea is basically you’re talking to it through language, and it’s abstracting pretty much everything else that we do at a lower level, that is like write the code, run it on a terminal, and see how the UI looks like, or the back end looks like, building API, things like that. So at the end of the day, we are all naturally going towards the state of abstraction and making sure the experience is solving the problem, but this problem is solving the back end.

So one of the things I wanted to ask was, now that you’ve been in this space for 20 years, tell me a little bit about what do you think are some of the common mistakes that companies are making when they’re choosing databases, and probably when they’re going through modernization effort and things like that?

Gajanan Chinchwadkar:

That is really a very controversial answer.

David Joy:

That’s fine. This is a controversial conversation.

Gajanan Chinchwadkar:

That’s a very controversial answer because there are some mistakes which companies cannot avoid. They’re almost forced to do those mistakes. That is because they have so much of legacy. Because of legacy, so I have seen some companies where they cannot use anything other than the VTU database. So in that case, telling them that you are making a mistake, in this case you should be using MongoDB, it doesn’t help, actually. So sometimes the legacy reasons influence these problems, sometimes policy or manageability also influence a database administration team is limited, and then that team can handle only three, four, or five such databases, not more than that. So all that is happening.

But typically one mistake that should be avoided is that, really trying to understand the application and the data architecture, and trying to match the data model for that application. If it is a really simple transactional data, and it is tabular data that you must use relational databases today, because not every database will give you that performance and that quality and consistency, asset compliance, all those properties.

So that is true. But at the same time, there are some problems which may be suitable for higher granularity, so you can use a document database, document oriented database. There may be some with smaller granularity that you could use, possibly key value stores or graph databases. So what is important is, are the relationships important? So after three years or four years of using this database, what is going to happen? Is my schema going to evolve? If my schema is evolving, then I should not use relational database. These kinds of fundamental questions people should be considering, and I think that doesn’t happen. So what they do is that typically somebody, some architect or some people in the DBA team familiar with particular databases, they say that you use this one, and then that one application starts on that database and then the database size becomes big. Once they accumulate a lot of data, then different business teams come and they say that we need different type of applications on the same data, but that cannot be supported with that data model. All that happened.

David Joy:

Yeah, II have seen those problems have before. I remember I’ve worked with a customer, a very popular customer in this tech space, and they started off by using DynamoDB, which it solved the problem really well for them, but suddenly its data size went to five petabytes and then the AWS bill, as well as performance, took a hit, and they were like, “Okay, now we need to get out this. What should we do?” And that is a paradigm at this problem keep happening. But some of the points you made definitely help. A tangent to that question I wanted to add was how do you feel about the evolution of the cloud and cloud infrastructure in this database space? Because there is a synergy in how data has grown and the cloud coming in to support that growth actually. And how have you seen that evolution in this space?

Gajanan Chinchwadkar:

It’s very interesting. I had a little stint at Amazon Aurora also, so-

David Joy:

Aurora. Wow.

Gajanan Chinchwadkar:

So it’s a cloud database. So I think cloud, it’s a big deal, basically it’s a great thing that happened, because people, especially for smaller applications, they don’t have to manage their own databases, they don’t have to take care of reliability or availability, those kinds of things, backups, restores, all that is taken care of by cloud databases. And so it is very powerful, in my view. And probably that is one of the reasons why so much of so many applications and hosted services are happening because managing databases, which used to be a headache in the past, it used to be a very expensive thing to manage databases, so people can now do that very easily and naturally because of the evolution of cloud, the whole application development ecosystem also kept evolving, and it became easier and easier.

So you can see so many services which a common man can use. In fact, if we just look at even our iPhone, or Android phone, you see how many services are available, and that too on one person’s phone. So all this is happening because of this cloud evolution. So it is a big deal. And for enterprises, I have also seen that sometimes they just want expansion. And it is very easy to just increase the size of the instance, and you can continue, your application continues functioning. In this world, serverless offering in databases that brings down the cost. So I think it is a very, very major development. In fact, even I think more personally, I think that Amazon Aurora was the original engineers who worked on that idea, they had done a great job. They observed very, very simple things in databases that almost everybody who works inside databases knows what is going on. They just observed that thing and just simply use that to split storage and compute.

David Joy:

Right. So you joined the Aurora team, were you part of the initial days of Aurora or were you once it was created?

Gajanan Chinchwadkar:

Yeah, much later. That’s why I appreciate those people who originally thought of this idea and those who implemented. I think that is like a beginning of the thought process for cloud databases. They separated the computer and storage.

David Joy:

Well, interesting point. I was curious to also understand from you, with what you’re seeing, the traditional role of a database administrator, do you think that role still exist as it is, or do you think it’s evolved into something else, or it’s going to stay the same?

Gajanan Chinchwadkar:

It depends on where the organizations are, because in certain organizations, the old one and the large ones, it still remains more or less the same. And there are many organizations where they are not on the cloud, so it remains same. But if the organization is on the cloud, then there is a little bit of DevOps kind of thing that is happening there. So it is very subjective, it is hard to characterize that or put that in one bucket.

David Joy:

Yeah, true. Yeah, I agree with you. I think nobody realizes this, that even though we are growing at such an active pace right now in terms of technology with AI and cloud infrastructure, a bunch of things, people are going to be surprised when they hear about the amount of Db2 mainframe that’s still running in the world, or Oracle running, because I think there is a bunch of activity going on in that space that is affecting businesses run their systems. So that’s going to be surprising for people when they get to do this, obviously.

So when it comes to trends, how do you personally keep up with all the innovation and R&D that’s happening? How does Gajanan decide to go, “Hey, I need to read this, read this, to keep up with what’s going on?” What’s your strategy? Tell it to the people. How do you keep up?

Gajanan Chinchwadkar:

It has become very, very crazy, because every day in my mailbox I get to see at least four or five papers recommended by many different experts. They’re expressing their opinions about those papers. But maybe sometime, few years back, it was not this active or continuous invention or continuous research kind of era. So the way I used to look at it’s always find out who are doing good work, particularly the professors who are doing good work, their students, and try to see what they are doing, what kind of papers they’re publishing. And out of those, even that used to be large number. So out of that, probably what does that interest me in my current work, or in my long-term interest, only try to read that much.

But generally because if you look at a conference proceeding, then you know roughly where the world is going, what people are thinking, what kind of problems they’re solving, that is the way how I used to go about it. And they’re spending time on the weekends to read papers, but it’s always outside of job, outside of work, this kind of thing. So that’s why it is a hard question because you really, if you had to keep up, you’ll really end up working 24/7.

David Joy:

That is true. I don’t disagree with you on that. I think there is definitely way more papers, especially research papers coming out, and I think there was a Twitter post I saw, X post, that said that about 85% of them is not really high quality breakthroughs and stuff like that. Of all the things that are happening in the space-

Gajanan Chinchwadkar:

For the AI papers also now, I became more and more careful. So nowadays I am following only top-notch researchers or top-notch research groups and their papers and their kind of work, probably that is good enough for a person.

David Joy:

I don’t know if you’ve been following, there’s a 1 million token limit now available in Google Gemini, and Claude released that recently too. So whenever I see a paper, I just drop it there and I do the lazy person’s reading of a paper, just to keep up. But I was curious to understand, obviously we talked about you reading papers and keeping up with things. What are the things that you are excited for going forward? Obviously AI is central to some of the things that you’re working on, but where do you think the tech world is going to be in 10 years? I know it’s a broad question, but being a veteran of the space, been in the industry for almost three decades, how do you feel like it’s going to look like in the next one?

Gajanan Chinchwadkar:

Sometimes it is scary to think about it, and I was myself discussing with some youngsters in my family that today somebody comes to me and asks, “Which computer science courses should I take?” Or, “Should I take computer science as a major?” I don’t know whether I have an answer for that, because it looks like lot of this work is going to change. I won’t say it will be eliminated, but it is going to change, and it may require a different mindset than the computer science mindset fully. Earlier software engineering was like, the better you are at fundamental foundations of computer science programming, it’s better. But today you can get 80% program developed by LLM, so you have to only correct fix it. And even sometimes you ask back to LLM that, “Can you correct this? This is not running correctly. These are the errors I’m getting.”

So with that, maybe programming will become less important, but problem solving may continue, at least in the near future. It doesn’t look like our competitors or AI is going to be able to think of different scenarios, different situations as much. And even if it does, [inaudible 00:34:55] computers do that, but it’ll not be applicable to all walks of life where software is required or software is used for which software is developed, basically. So I think that particular skill will still be in demand. So I would say, recommend somebody going new in this field, of course be good at mathematics, for sure, and then try to learn how to learn fast. Basically, how to learn. That skill is more important than just learning.

David Joy:

People don’t understand this fact that your ability to learn things quickly is basically a skill in itself. Not just the fact that you can do something faster or quicker than anybody else, but just that you absorb things quickly and you can apply something quickly. It’s a different way to operate, actually, than so many different people. I’ve met some people who I’ve seen, will you show them a white paper, and they are able to digest that in 10 minutes without using an LLM, and then they can come back and say, “Oh, well this is about this.” It’s just a different skill altogether. Anyway, so I know we have to be cognizant of the time, Gajanan, and thank you so much for coming on and talking to me about some of the things. Tell me, where can people follow you and Hypermode? What do you want people to know about Hypermode? What do you want people to know about you? And where can they find you guys?

Gajanan Chinchwadkar:

Of course, they can find me on LinkedIn. That is one thing. And Hypermode, we do have our own website now, company, although the company is recent, but we do have our website, and it is evolving constantly so people will be getting newer and newer things, we’ll be seeing newer things on our website. And it will be interesting to get someone who is interested in solving AI problem and they do not know where to start, it’ll be interesting to chat with them and enable them to solve applications using AI, using Hypermode, I’ll be very happy to do that. And it’s very easy, my email address is g@hypermode.com.

David Joy:

Well, that’s a great, great email, just G.

Gajanan Chinchwadkar:

Yes, g@hypermode.com.

David Joy:

I just wanted to say, Gajanan, thank you so much for taking the time to come on the podcast. It’s been an absolute pleasure. For everybody listening in, go check out Gajanan on LinkedIn, as he was saying, and if you want to reach out to him, hit him up on his email that he just shared. It’s been an absolute pleasure talking to you, and thank you so much for sharing your experience with so much humility. I know I’m pretty sure you have done so many amazing things in this space, we could have gone into so many places, but you kept it grounded, and I appreciate you taking the time to come on.

Gajanan Chinchwadkar:

Hey, thank you for inviting me here, and for this nice opportunity to interact with you and also learn from your experiences. Thanks.