Science
Reasoning, retrieval, design, and in-silico verification joined into one auditable loop — every conclusion traceable to its evidence and to the code that produced it.
The loop
Every step inspectable, challengeable, reproducible — not a black box.
Reason
Multi-step reasoning you can inspect, challenge, and cite.
Retrieve
Evidence synthesized across 66,000+ full-text papers and a knowledge graph, with provenance.
Design
Sequences produced by verifiable tools and checked for validity — correct by construction, not hallucinated biology.
Verify structureLive
Designs are structure-predicted (Boltz-2), colored by confidence (pLDDT), and emit a reproducible structure artifact.
Auditable artifactsIn dev
A reviewer agent checks citations and numbers; figures, structures, and claims carry their code and provenance.
Structure verification
A real verify_structure call: the same sequence, well-folded vs scrambled — Boltz-2 predicts pLDDT 0.93 vs 0.48, an in-silico structure-confidence screen that separates plausible from suspicious folds. It is a predicted-confidence signal, not experimental validation of folding, expression, or yield.
Structure verification is live; the reviewer and full-loop orchestration are in development.
Where Evolrix AI sits
AI is turning the cell into something you can reason over and verify. Evolrix AI brings that approach — reasoning + domain tools + in-silico verification in a closed loop — to one practical industrial problem: expressing proteins well in Pichia (Komagataella phaffii). We are not building a general virtual cell; we build the industrial-protein-production slice: structure verification, expression-construct recommendations, and literature & patent evidence, joined into one auditable workflow.