Healthcare AI stewardship prototype using FHIR context, MCP tools, and A2A/ADK agent concepts — research/hackathon framing, not a production clinical system.
Project Autopsy
Same project, six engineering perspectives. Switching lenses keeps RUN and X-RAY mounted so lab/scrubber state is not wiped.
Healthcare AI stewardship prototype using FHIR context, MCP tools, and A2A/ADK agent concepts — research/hackathon framing, not a production clinical system.
Deterministic Runtime Lab — PORTFOLIO_EXTENSION. Prefer this lens over a separate route when you want lab state to survive inspection switches.
Synthetic patient A (demo bundle shaped)
States · complete 5 · warning 0 · blocked 1 · error 0
| Tool | State | Detail |
|---|---|---|
| get_active_antibiotics | complete | Synthetic active antibiotic list loaded for inspection. |
| get_culture_results | complete | Synthetic culture / sensitivity profile loaded. |
| check_patient_allergies | complete | Allergy cross-check ran on synthetic records. |
| get_renal_function | complete | Synthetic renal labs present for dose-context inspection. |
| generate_stewardship_recommendation | blocked | Recommendation generation intentionally blocked — this lab withholds treatment advice. |
| create_stewardship_task | complete | Dry-run FHIR Task preview only — nothing is posted to a server. |
Baseline path proves tool orchestration with complete synthetic context. Treatment advice is withheld; Task remains a dry-run preview.
Treatment advice withheld by design.
Project X-Ray — responsibility layers with isolation, component selection, and Source Trace. Causal reconstruction (M9) remains below.
Steward Project X-Ray / Reversible Architecture deferred — Agent Lab RUN is the M12 inspection surface. Prototype status stays visible on STORY / DECISIONS / EVIDENCE.
Engineering decisions grounded in public evidence — not a fabricated post-mortem diary.
Public MCP server surfaces FHIR-oriented tools (antibiotics, cultures, allergies, renal, recommendation, Task) so orchestration can be reviewed step-by-step.
README disclaimer frames Steward_AI as research/education — not medical advice and not a production clinical system.
The inspectable lab uses synthetic context and withholds treatment advice. It is not live FHIR, Gemini, or MCP production runtime.
Exhibits appear here only when they clear the evidence gate. Future renderer is ready; the canonical dataset is currently empty.
Research/hackathon prototype — not medical advice; not a production clinical system.
Deterministic portfolio simulation of stewardship tool states on synthetic FHIR-shaped context — not live FHIR, Gemini, or MCP production runtime.
PORTFOLIO_EXTENSION — browser Steward lab grounded in public Steward_AI MCP/README evidence; synthetic context only; not live FHIR/Gemini/MCP.