Skip to content

README.md

Latest commit

 

History

History
25 lines (17 loc) · 3.4 KB

README.md

File metadata and controls

25 lines (17 loc) · 3.4 KB
Raw

real_data/ — verified, licensed sample data

Optional, real/synthetic sample data to make SEPHIROTH's demo experience realistic — patients with actual clinical histories, real drug-interaction severities, and a real chest X-ray, instead of only the two hand-typed demo patients. Nothing here is required to run the app or the test suite — see What doesn't change below.

Each source was chosen after explicit research into which one is genuinely verifiable and appropriately licensed for this use — see the conversation history / project docs for the comparison against alternatives that were rejected (NIH ChestX-ray14, CheXpert, MIMIC, RxNav, openFDA, DrugBank, n2c2, etc.) and why.

Folder Source License Committed? Refresh command
patients/ Synthea Coherent Data Set (MITRE) CC BY 4.0 / Apache 2.0 ✅ sample_patients.json (12 patients, parsed) python3 real_data/patients/fetch_synthea_sample.py
notes/ Derived from the Synthea patients above, via SEPHIROTH's own Gemini client (same role as synthetichealth/chatty-notes) 100% synthetic ✅ sample_notes.json (6 notes — see its README for an honest note on how these were actually authored) python3 real_data/notes/generate_notes.py
drug_interactions/ DDInter 2.0 CC BY-NC (non-commercial) ✅ ddinter_subset.json (193 pairs, severity real + tier-based advisory text) python3 real_data/drug_interactions/fetch_ddinter.py
imaging/ RSNA Pneumonia Detection Challenge (via Kaggle) Academic/non-commercial, not redistributable ❌ never — samples/ is gitignored python3 real_data/imaging/fetch_rsna_samples.py --count 10 (needs your own Kaggle credentials)

Each subfolder has its own README.md with the full detail on what's actually in the data, its exact license terms, and how it's used by the app.

What doesn't change (and why)

  • platform/core/db.py::SEED_PATIENTS (the two demo patients, P001/P002) is untouched — it's the deterministic fixture the test suite depends on. Synthea patients are additive, imported separately, never auto-seeded.
  • intelligence/mcp/drug_safety_server.py::INTERACTIONS (hand-curated pairs) is untouched and always takes priority — DDInter only fills gaps, and gracefully degrades to hand-curated-only if ddinter_subset.json is ever missing.
  • Nothing in this folder runs automatically — not in pytest, not in CI, not in init_db(). Every fetch_*.py/import_*.py/generate_*.py script is meant to be run by hand, exactly like the existing data/embeddings/build_artifact.py and the references/ clone instructions in the main README.
  • data/rag/SEED_GUIDELINES (the RAG corpus) is untouched — no programmatic source was found reliable enough to expand it automatically (see drug_interactions/README.md and the project's evaluation research); it keeps growing via the /add-guideline skill.

Licensing summary (read before any commercial use)

If SEPHIROTH is ever monetized, drug_interactions/ (DDInter, CC BY-NC) and imaging/ (RSNA, academic/non-commercial) both require a separate commercial license or removal before that happens. patients/ and notes/ (Synthea-derived) have no such restriction.