Lucas E. Brennan-Almaraz
Stanford CS. Building AI infrastructure for biology at Arc Institute.
Publications
Aditi T. Merchant*, Daniel Guo*, Ben Viggiano*, Lucas Brennan-Almaraz, Evelyn Hur, Tina Mai, Peter Yin, Samuel H. King, Euan A. Ashley, Brian L. Hie
Proto is a high-level programming language for generative biology that composes a small set of abstract primitives into structured programs, encoding generative design campaigns across DNA, RNA, proteins, ligands, and their interactions. It folds predictive models into generative workflows — used here to design alternatively spliced introns validated in human cell lines — and is natively multi-objective, reaching leading experimental success rates for synthetic protein–DNA design. Paired with AI agents, it lets users specify complex pathways and regulatory logic through natural-language instructions.
Projects
Experts play nothing like Nash yet score like it — the payoff matrix is effectively rank-3, so convention is free. Offline RL just finds a different convention, not a better strategy.
Players switch strategy far more after a loss than a win — yet that post-loss signal actively misleads prediction. Sequential transformers over 382K VGC replays.
Fetch any web page and return clean, LLM-ready markdown. Handles anti-bot protection automatically.
Python __all__ chain validator and pre-commit hook for enforcing clean public APIs.
Phone-to-3D indoor reconstruction from a handful of iPhone photos, compressed to sub-8-bit for mobile deployment. Highlighted on the CS231n Spring 2025 reports page.
First successful integration of LoRA fine-tuning with the Mamba architecture.