A summer internship reflection on autonomy, hard problems, and building across the Overmind product.
Who knew spilling Thai food down my future boss’s pants would give me a summer internship? Well, that’s what happened to me. I was at a dinner in San Jose during my trip to NVIDIA GTC, where I met Sam, my future boss. Upon exchanging ideas about the ever-evolving landscape of agents and LLMs, we seemed to agree upon one thing: agents are nowhere near perfect and have so much room for improvement, whether in cost, reliability, speed, or efficiency.
My internship at Overmind has been an unconventional one. In a good way.
It genuinely feels like I just joined last week - time really flies when you are having fun - but it simultaneously feels like I joined a year ago. So much has happened since joining, which is truly the magic of working in a startup.
Most internships are structured in a fairly common way: interns get assigned to a mentor or supervisor, they get assigned a project, and they work that project to completion throughout their time at a company. Things were different here. From the week I joined, I was able to contribute to the core product. It felt extremely empowering, but I also remember asking the Founding Engineers an awful lot of questions. Sorry.
What was the culture like?
From the get-go, I was given a lot of autonomy and was involved in core discussions around the product. This also meant that there were a lot of hard problems to solve. I genuinely liked the fact that Overmind encourages employees to come to the office. There’s just truly something about collectively staring at a whiteboard in the office and brainstorming together. We were able to tackle the issues and bounce ideas off each other, and it felt like we had each other’s backs.
What did I work on?
I pretty much worked on all product surfaces. From evals and agentic behavioral analysis to model fine-tuning and harness optimization, the whole pipeline is connected. I was given the space and bandwidth to experiment and come up with my own solutions to problems. Sometimes, I’d be working on something from the ground up, and sometimes I would be collaborating with someone else. Sometimes I was doing hardcore engineering, and sometimes I was doing UI/UX reviews or running marketing case study experiments. Given that Overmind is relatively early-stage, it really felt like I was deeply embedded in the roots of product development from scratch, which felt eye-opening, since I was exposed to so many facets of building a real customer-facing product.
What was so eye-opening about it?
As a Computer Science student, we are often only exposed to the technical side of things, like engineering-related concepts. However, there truly is so much more to building something that solves a gaping void in the market. The whole element of whether people would actually use it, how to market your product, how to get people hooked onto your product, and how to grow your brand and image in a vast and competitive agentic space - all these questions do not have easy answers, and they are probably not something you could conventionally comprehend in a lecture hall. In this regard, I am truly grateful Overmind has opened my eyes.
Want to work at Overmind?
The most important requirement: you need to like cookies and Filipino food. Don’t ask me why.
Just kidding.
Overmind really looks for people with a drive to solve the hardest problems in the agentic space and truly understand the gaps that need to be filled from the messy noise out there. Everyone’s building in the AI space now, but agents are still riddled with endless problems. You need to be someone who is willing to dive deep into problems, and most importantly, adaptable enough to venture into uncharted territory.
To end off
It was really fun exploring so many different areas of engineering, like data science, classical ML, software engineering, model fine-tuning, and agentic engineering. I managed to learn so much in a relatively short span. These were things you simply can’t learn in a lecture or classroom. As such, things will need to be self-taught, and a growth mindset in a startup environment like Overmind would prove vital.
That said, there still genuinely are so many problems in this space, from evals to pre-training pipelines to training and post-training backtesting. The whole model development cycle needs to be streamlined, and that’s truly what Overmind is all about.
