From the Cave to the Customer: Lessons from My First 100 Days at Groq

By Lev Zlotnik , Machine Learning Development Specialist

My name is Lev Zlotnik and I’m a Machine Learning Development Specialist at Groq. My focus is helping customers achieve real-time AI - running their workloads on GroqChip? to realize the fastest time-to-solution with the single push of a button. That’s the vision; and from what I’ve been able to see and do at Groq, I’m convinced reality can and will keep up.

After spending the first part of my career in the Israeli Air Force and at Intel, joining Groq was exciting because it gave me the opportunity to work in a hands-on environment and interact with customers, a welcome change from working behind the scenes solely on software technology. I knew by joining Groq I’d be working with some really smart people solving some really tough problems.

In my first one hundred days at Groq, I had the opportunity to use two of my favorite tools, math and physics, to tackle machine learning challenges, working on projects ranging from fusion reactors to advertising recommendation systems to autonomous vehicles. Here are some fun surprises and valuable lessons I’ve learned along the way.

Work where curiosity is valued - this is where you will have impact?

When given a customer project to support, I usually start with understanding the problem they’re solving and - if they’re willing to share it - their algorithm or another proxy to their approach, so I can then analyze it and understand how it will run on GroqChip. I look at every problem as an opportunity to explore a new domain, which includes research, reading academic papers to learn about the problem and the approach, and investigating the mathematical functions modeled to understand how they produce certain behaviors.?

I feel lucky that my preferred process is very similar to the way Groq approaches problems: looking at things from first principles. It’s about having the curiosity to explore and wanting to solve a problem while limiting assumptions, always asking why, and trying to understand problems from the inside out to reach the best solution. This all means I don’t just work on passing neural networks through Groq Compiler - I research math in-depth to understand how a concept applies simultaneously to both a customer’s application and our technology.

I also work closely with the software leads, which has been a really positive experience. They’re very supportive of new problem solving approaches, and appreciate that I try to generalize problems to the widest range possible and then solve them from their root cause.

After just a few short months of working with Lev, Brian Kurtz , Director of Design Acceleration at Groq, observed, “While some people are really focused on accelerating the piece of a workload that’s the problem, Lev looks to treat the problem, not just the symptom. He did this recently for one of our autonomous vehicle customers. By understanding the problem holistically, he’s helping develop broadly useful tools that are having impact in multiple places."?

“We” delivers more than “me”

I originally joined Groq because I wanted to work at a start-up company where my work would make an impact. I’m technically in the sales org, which may surprise fellow software engineers, but I’m always collaborating with the engineering team. Both groups understand how to make a workload run on GroqChip and I try to be the customer’s advocate as we problem solve. Sometimes we do this by “pair programming” which means teaming up with someone else who has a different area of expertise, say hardware, and working together on one set of code to make it work.

This team-centered culture pushes us to combine architecture and engineering parameters with customer requirements into solutions that really impress our customers. And it’s a big part of what makes me feel like I have a lot of impact at Groq.

Kyle Aubrey , Head of Customer Apps at Groq, said, “Since joining my team, Lev’s not only helped customers achieve amazing performance with their models, he is also constantly working with our engineering teams to drive improvements to our development tools and examples. His contributions have helped us establish high speed developer velocity as a cornerstone of Groq products and software.”

The era of deterministic compute has arrived

Coming into Groq, I was most excited about having an impact with my existing knowledge, but I was immediately humbled by how different their approach to computing is. I didn’t expect it but something I’ve enjoyed most has been modifying my pre-existing notions with what I now think is the proper way of simplifying compute.?

From day one at Groq, I’ve learned by doing. All new hires on the sales and applications team are given the chance to run a workload on GroqChip that they are already familiar with. I decided to start with a simple workload and then moved on to experimenting with an application-specific workload. I then decided to try to generalize it and I ended up getting it working in time to showcase to a customer who needed real-time AI to solve their challenge. The customer then handed us another workload on a Thursday and we had it working before the weekend.

Instead of just hearing how different our technology was, I got to get my hands dirty by exploring and helping customers. This all happened in my first two months at Groq and it’s blown me away to be working where I’m not held back and am enabled by an entirely new approach to compute and the amazing efforts and helpfulness of my colleagues.

Read this and more Groq insights here . If you’re interested in joining us, explore Groq careers .

First photo credit,?Dave Ashutosh ?on?Unsplash

Second photo credit, Adobe Stock

Third photo credit,?Martin Sanchez ?on?Unsplash

Fourth photo credit,?Johannes Groll ?on?Unsplash