Find
Read a patient's genome to determine which existing therapies will work, which will fail, and which will cause harm — before the first dose is given.
Candidate matching against an individual variant profile
Design
Generate entirely new molecules against targets that have no drug. Not screened from a library — assembled atom by atom against the specific surface in front of us.
Computational protein design — from sequence to structure
Edit
Correct the defect at its source. For inherited disease the problem is not a pathway to suppress — it is a sequence to rewrite, once.
Edit site, insert, and re-formed sequence — designed as one
Command
Direct biological systems rather than merely inhibiting them. Instruct immune cells and cellular machinery toward a chosen outcome instead of blocking one node and hoping.
Models we train, composed with open science we run.
Part of the stack we train ourselves on data we compile and curate. The rest are open scientific tools we license and run — Boltz-2 and GEMS (both MIT), DiffSBDD, and ProteinMPNN among them. The engine is the composition: how those models are chained, constrained, and scored against one another, and the outcome data that comes back.
Three layers. Each one makes us harder to replicate.
Any single layer can be rebuilt by a well-funded competitor. Rebuilding all three — and the compounding between them — is the actual barrier.
Biology Moat
Patient outcome data flows back into our models and nowhere else. Every treated patient makes the next design better — an advantage that cannot be bought, only accumulated, and that a newer entrant cannot start with.
AI Moat
Models spanning target identification, generative design, structure prediction, and safety — the ones we train sitting alongside open tools we license and run. The advantage is the composition and the shared constraints, so the system reasons across objectives instead of handing files between them.
Manufacturing Moat
A design only matters if the molecule can be made. Manufacturability is a constraint inside the engine from the first candidate, and specifications go to our CDMO network sealed — production scales without the design leaving our control.