Engines¶
The horizontal compute muscle of the factory. Each engine is a registered capability with a typed interface and an honest status.

Engines compute; agents interpret. An engine produces a result that can be reproduced without a language model — a variant call, a cluster assignment, a docking score. The intelligence agents reason over those results and return cited clinical context. Keeping the two separate is what lets each be checked independently: the numbers reproduce on their own, and an interpretation can always be traced back to the numbers it rests on.
The eight engines are horizontal — they serve any condition. A disease program is a vertical that composes them for one condition, and is not a ninth engine.
| Capability | What it does | Domain | Status |
|---|---|---|---|
| Cardiology Engine | 11 clinical workflows and 6 risk calculators spanning prevention, intervention, and rhythm; CAC + 10-year ASCVD prevention logic (ACC/AHA), handing to… | clinical | live |
| Clinical Imaging Engine | Live DICOM analysis on real pixels: the live-verified modality is chest X-ray via DenseNet-121, with FHIR R4 export and cross-modality reasoning against genomics ⚠️ stated limits — see the capability page | clinical | live |
| Genomics Foundations Engine | GPU germline variant calling: FASTQ to BAM to VCF via accelerated alignment (fq2bam/BWA-MEM2) and DeepVariant ⚠️ stated limits — see the capability page | genomics | live |
| Large-Molecule / Structural Biology Engine | Large-molecule / structural-biology engine (E7): real protein structure prediction and design served on the GB10 — ESMFold folding, ESM-2 embeddings +… ⚠️ stated limits — see the capability page | proteins | planned |
| Precision Intelligence Engine | Variant annotation (ClinVar/AlphaMissense) + RAG interpretation over the shared variant foundation to identify and contextualize druggable targets | clinical | live |
| Precision Oncology Engine | MTB packet generation, therapy ranking, and trial matching, with cross-modal joins to the imaging, biomarker, and trial capabilities; fusion-first pediatric MTB,… | clinical | live |
| Single-Cell Analysis Engine | Engine 8 compute layer. Real scanpy scRNA-seq workflow (QC -> normalize -> HVG -> PCA -> Leiden clustering -> marker DE) + marker-based cell-type annotation | single_cell | verified |
| Therapeutic Discovery Engine | Target to ranked drug candidates: generation, docking, and RDKit scoring across a 10-stage pipeline ⚠️ stated limits — see the capability page | small_molecules | live |
How to read this table
Every row — including the status — is generated from the capability registry, the same file continuous integration validates. A capability marked live is served by a real model against real input, never a mock; anything planned is labelled and not shown as if it ships. See the full Capability Maturity Matrix and the Honesty Ledger.
All clinical output across these capabilities is decision support for a qualified clinician — never autonomous diagnosis or prescribing.