I build ML systems and run experiments on how language models behave.
I graduated from Yale with a CS degree in May 2025, after writing my senior thesis on evaluating language models. These days I direct LabOS, an agent-driven ML research lab that explores my ideas in parallel, and run a profitable trading system on the side. Before that I was co-founder and CTO of Axis, a Y Combinator–backed startup.
Startup & trading
Other projects →A Y Combinator–backed startup building an agent-assisted research dashboard for commodity markets: ingestion pipeline, provenance-tracked knowledge graph, forecasting backend — agents do the legwork, analysts stay in the loop.
A fully automated market-making engine on prediction markets — order-book ingestion, quoting, inventory control, and operator-acknowledged incident handling across two live accounts.
Research
Inside LabOS →Splice a re-encoded edit into a stale KV cache and the repair is exact except after the edit. That residue never flipped an unrelated query (0/252, twice) — and flipped a third of the queries that read the edit.
A translator-free lens for the layer where a transformer settles each token, validated on three model families. It out-tracked the trained tuned lens — and showed the depth you read off depends on the ruler.
Can LLMs predict how people answer survey questions? An evaluation suite with distributional baselines and a dummy-option ablation that separates learning the format from learning the behavior.
How gradient-based optimizers hold up when a fraction of the training data is adversarially corrupted — theory, simulations across contamination levels, and a hybrid of the two methods.
Tagged LabOS = work from my agent-driven research lab: experiments run by AI agents under my direction.
Before Axis, I was a junior quant strategist at Akuna Capital and a quant trading intern at Virtu Financial, did research in LLM evaluation and condensed-matter physics, and built a social app that reached 500+ users. More about me →