Scientific AI
AI that works through the biology—not around it.
We are developing AI workflows that use structured immunology knowledge, scientific evidence and inspectable mechanisms to help people explore questions and organise research. Outputs are limited by the available graph, evidence and validation.
How we talk about AI safely
- Prefer AI-assisted scientific exploration over “superintelligent discovery.”
- Model-generated ideas are candidate hypotheses — show alternatives and disconfirming evidence where available.
- Show graph release, retrieved relationships, source IDs and tool versions where available.
- We do not claim clinical validation, experimental confirmation or superior discovery performance without measured evidence.
Capability status
- Ask (graph-retrieved teaching) PartialAsk retrieves graph context and reviewed relationship cites when present — not fully evidence-grounded AI.
- Literature intelligence PartialLocal candidate claim workflow and curation export; no automated bulk literature ingest.
- Omics / molecular AI adapters PlannedInterface stubs return not_available until a real tested adapter ships.
- Scientific research workbench PartialGuided question → evidence → mechanism → hypothesis → experiment → review; candidates only.