Transformation of the Week
Akash Shetty

Transcript

Lusk: Hey Akash, thanks for joining me today and congrats on winning the Transformation of The Week. Tell me a little bit more about your background and what you’re doing today.

Akash: Hey Chris, thanks for having me. My background is, I’m a physicist and mathematician by training, but I’ve been working in AI, or data science, since 2017. Since then I’ve gone to work as a data scientist and machine learning engineer, and now working as a CTO at Publicus.

Lusk: So what first got you interested in Gen AI?

Akash: LLMs I would say since 2018 or 2019. So we used to use the early BERT models, the GPT models. I used to work at a company called Wonder AI. So we used to look at a lot of our documents and data, so we used to use it back then, but it was not at the quality of like we have right now. And then I was working mostly on search and recommendation problems. In these use cases we often use language models, or LLMs for embeddings for improving vector search and just search recommendation itself. That’s where we started by using it specifically for embeddings and a little bit of generation stuff. But I think since 2022 it just blew up, it just got so good that like, undeniably you’ve got to use it.

Lusk: Now, you graduated from our AI Engineering Bootcamp cohort number four, a couple of months ago. Have you been able to put anything that you learned into practice, whether it’s at work or on personal projects?

Akash: AIE 4 was really good for me. I got a lot of experience working with really cool people like Greg and Wiz, and I had just quit my job, and I was working on my startup, and I met my co-founder here. Since AIE 4 I think me and my co-founder Joe, who is also based in Toronto, we bonded quite well. And we’ve been building Publicus from the get go since mid program. So I think the, final project of the course was also something related to Publicus but now we’ve gone heavy into building a software product and AI-based applications on our product.

Lusk: So let’s dive a little bit deeper into your startup, Publicus. What are some big AI challenges that you and your team are tackling right now?

Akash: I think the rapid development of AI has been a big challenge in the sense that, I used to try focusing on building multi-RAG systems, like configuring different kind of embedding or chunking strategies and so on. But out of the box models are really good, models like anthropic and Claude have PDF support natively. So a lot of features that we’ve been building is kind of redundant and cheaper.

And second is, trying to understand the ceiling. These LLM models are really, really good, but they do have their fallbacks and identifying the fallbacks, not now but where we’ll be like six months or maybe 12 months down the road, would help us understand better on what tasks we need to work purely with LLM-based agents or applications. And what tasks do we need to use more traditional based approach? I think that hybrid is the thing that we want to focus on, but it’s still something we’re figuring out.

And I think the last piece is product development has changed. We have to use AI to develop our product. Understanding how to build a company with AI first philosophy in like, anything we do in terms of operation. I think we need to ensure that AI is part of it because it just expedites things much faster.

Lusk: So what words of wisdom do you have for somebody out there that’s watching this who is considering a move into Gen AI?

Akash: It is a hyped time right now, right so everybody’s using it. So what I would say is let’s get our hands dirty. We keep hearing that Gen AI saves time. So pick anything that you think you want to save time on, or you don’t enjoy doing, and it’s repetitive; let’s do a gen AI.

I’ll give you an example that I’m trying to do myself is I don’t have a job. I’m working at my own startup, so I don’t make much money. So I need to figure out my food habits and eating habits. How I use Gen AI for this? I basically take photos of every item I have in the fridge, all my pantries and spices. Then I check out a bunch of fliers which have deals, and I use basic ChatGPT and Claude to prepare a recipe based on what’s available in my kitchen, what deals we have, and what I taste preferences are.

So I think finding quick use cases of using different out-of-the-box solutions and finding an application, is the way to go.

Lusk: That’s great advice! You are the epitome of build ship and share. Congrats again on winning the Transformation of The Week, tell the people out there how they can connect with you.

Akash: I think the best way to connect with me would be LinkedIn and X (Twitter).