Skip to content

Open-source precision medicine · Apache-2.0

No one should wait years for a disease we could have understood in a day.

Patient DNA to a shortlist of therapeutic candidates, in hours — on one box you can fork and run yourself, bursting heavier work to remote GPUs only when it needs to.

Explore the factory Run it yourself

Decision support for a qualified clinician — never autonomous diagnosis.

hoursnot months
8 · 8 · 1engines · agents · program
9verified on real data
Apache-2.0commercial use welcome

The pipeline

From patient DNA to therapeutic candidates — in hours.

Raw DNA to variants to interpretation to a shortlist of therapeutic candidates

Illustrative — hours, not months. See the demonstrations →

Home

The three-minute tour

A narrated, captioned walkthrough of the whole factory — the eight engines, eight agents, and the TSC flagship. Decision support for a qualified clinician, never diagnosis.

Inside the factory

In one line: the factory takes a patient's DNA and helps a clinician get from raw data to a shortlist of possible treatments — in hours, not months. Under the hood it's eight compute engines, eight clinical-reasoning agents, and one flagship disease program, on a single NVIDIA DGX Spark (a desktop-sized AI computer), open under Apache-2.0, bursting heavier work to remote GPUs only when it needs to.

  • 8 Engines

    The horizontal compute muscle — genomics, clinical interpretation, therapeutic discovery, imaging, oncology, cardiology, large-molecule / structural biology, and single-cell.

  • 8 Intelligence Agents

    The clinical reasoning layers that interpret what the engines produce — each one decision support for a qualified clinician.

  • 1 Disease Program

    Tuberous Sclerosis Complex — the flagship vertical that composes the whole factory for one child, in one governed afternoon.

Explore

  • The Factory — the Platform → Engine → Agent → Program model, with a registry-generated page for every capability.
  • Capability Maturity Matrix — every capability's real, audited status, generated from the registry. The site cannot claim ahead of the code.
  • Run It Yourself — clone it, run the gate, verify a real capability.
  • Demonstrations — the D1–D7 portfolio, each with its honesty flags.
  • Capability Brief — the full picture in a technical cut and a mission cut.
  • Source on GitHub — read the code, fork it, open an issue.

What it is — honestly

The honesty discipline is the load-bearing wall of this project, so the limits are stated plainly everywhere they matter:

  • Decision support, not diagnosis. Every clinical output supports a qualified clinician — never autonomous diagnosis or prescribing.
  • A bench, not a cure. Where the factory points toward gene therapies, those are preclinical research — the promise of a direction, not a treatment available today.
  • "Elastic burst," not "all on one box." Heavy or ARM-incompatible models burst to remote GPU over a private mesh; nothing pretends to fit on the desk that doesn't.
  • Real, never mocked. A capability marked live is served by a real model against real input — and a code-level audit keeps that claim earned.

Source of truth

This site reflects the canonical roster — 8 engines · 8 intelligence agents · 1 disease program (TSC) — generated from the capability registry (lib/hcls_common/capabilities.json).