We transform black-box foundation models into transparent tools for drug discovery. Our interpretability platform reveals the biological mechanisms behind AI predictions, helping you validate drug candidates and de-risk your pipeline.
We use cutting-edge AI methods to decode how foundation models understand biology.
We decode the AI's internal representations to identify specific biological concepts it has learned, like cell types and disease pathways.
We map how the model connects these concepts to make predictions, revealing the biological mechanism behind each drug's effects.
We trained sparse autoencoders on the Tahoe-x1 model to decompose its learned representations into interpretable biological features, revealing how the model encodes cell types, pathways, and disease states.
CEO
Stephen leads computational biology and genomics development, and leads business relations. He is a PhD candidate at UC Berkeley focused on AI applied to biology.
Website →
CTO & CSO
Thomas leads mechanistic interpretability, causal inference, and core AI research. He is a PhD student with Yoshua Bengio and Guillaume Lajoie at Mila, and former research fellow at Anthropic.
Website →