Thesis
The AI discovery engine.
Thesis is an applied AI lab accelerating the frontier of AI discovery.
We're making it possible for researchers anywhere to discover the next Transformer or invent the next AlphaFold.
Mission
Our mission is to make state-of-the-art AI discovery 10x faster and 10x cheaper. The big labs are racing to get there, but their need to productize and turn a profit has already weighed them down. Thesis embraces rapid experimentation and novel fundamental research. Our singular focus, outcome-driven mindset, and early results give us confidence that we can get there first.
Results
Thesis is state-of-the-art on OpenAI’s Machine Learning Engineering benchmark (MLE-Bench).
MLE-Bench tests how well AI systems can train ML models autonomously, a first step toward self-improving systems. We achieved this result in the record time of 1 month and with only $10k in compute. Agents will get you far, but not far enough. Our methods mix old school with new, and have already outperformed teams of researchers at Microsoft, Google, Meta, and Baidu.
Team
Luigi and Sergio are brothers. Back in 2021, Sergio wrote about a future where AI could create its own algorithms. With today's foundation models, that future is suddenly within reach.
Sergio worked on AI R&D at Google X, Nvidia, and Stanford’s AI Lab (SAIL). He has published work at NeurIPS and ICML, and has worked with Chelsea Finn and Andrew Ng.
Luigi built Sphere, a fintech company that processes billions of dollars every year in global cross-border payments. Sphere was recently valued at $250 M+ in its Series A round.
10 Year Vision
The holy grail of AI is building self-improving systems that, guided by human creativity, enable breakthroughs and new paradigms achieving state-of-the-art results under resource constraints.
In the limit, our engine will expand beyond ML to every domain of science, from computational biology, to theoretical mathematics and physics. Thesis will autonomously discover new knowledge and become the driving force behind humanity’s scientific progress.
Our Ask
We collaborate with leading institutions, startups, and independent researchers around the world. If you’re interested in what we’re building, please get in touch: [ask@thesislabs.ai](mailto:ask@thesislabs.ai)
How did your company get started? (i.e., How did the founders meet? How did you come up with the idea? How did you decide to be a founder?)
We are brothers, best friends since birth. We grew up in the Caribbean with our mom, did math on small islands, taught ourselves calculus from MIT OpenCourseware when we were 12, and published graduate math by the time we were 16. We built worlds together in Minecraft as kids; now, we’re building the next generation of scientific computing.
The idea for Thesis clicked when we realized there was no agentic tool for computational work: nothing that could autonomously reason through complex data, whether in science, AI, finance, or any data-driven domain. Once we saw that gap, the path was obvious. We’ve always wanted to build things together and chart our own future, so becoming founders felt like a natural extension of who we are.
How did you decide to apply to Y Combinator? What was your experience applying, going through the batch, and fundraising at demo day?
Sergio applied to Y Combinator while working at Google X. Across his time in AI R&D at Google X, Nvidia, and Stanford’s AI Lab, he kept running into the same gap: researchers exploring data, running experiments, and building models had no real tooling: nothing like the workflows available to regular software engineers.
That insight sparked Thesis. As Sergio and Luigi began exploring the idea together, it quickly became clear that what started small had far larger implications. Luigi had been scaling his last company to billions in cross-border payments, but the chance to work on something that could fundamentally reshape how science advances was far more compelling. If we executed well, Thesis wouldn’t just be another tool: it would become the engine for scientific discovery itself. That conviction ultimately shaped our experience through the batch and beyond.
What's the history of your company from getting started until the present day? What were the big inflection points?
We’ve just begun.
What is the core problem you are solving? Why is this a big problem? What made you decide to work on it?
Computational science still runs on tools built for a pre-AI world. Software where researchers analyze data and make discoveries, were never designed for automation. Today’s coding agents can write code, but don’t understand the data they’re acting on. Scientists end up doing manual, fragile, error-prone work. We felt this pain firsthand across our time in AI R&D and saw the same gap everywhere: researchers have no infrastructure to automate experimentation or extend their capabilities. Thesis solves this by bringing data-aware AI agents directly into the scientific environment: virtual teammates who monitor experiments, analyze results, surface insights from the literature, and run background workflows safely and reliably, turning discovery into an assisted and continuously improving process.
What is your long-term vision? If you truly succeed, what will be different about the world?
Thesis will autonomously discover new knowledge, becoming the engine of humanity’s scientific progress. In the limit, Thesis will generate ideas, run experiments, interpret results, and iterate. Every major breakthrough, from new materials to new energy systems, will be invented on Thesis.
