https://github.com/peterkc/health-data-platform
AI-native workflow and governance layer for healthcare verticals — ambient capture, structured assessment (OASIS-E), human-in-the-loop review, EMR write-back. Home health first.
https://github.com/peterkc/health-data-platform
fastapi fhir healthcare home-health human-in-the-loop python
Last synced: about 3 hours ago
JSON representation
AI-native workflow and governance layer for healthcare verticals — ambient capture, structured assessment (OASIS-E), human-in-the-loop review, EMR write-back. Home health first.
- Host: GitHub
- URL: https://github.com/peterkc/health-data-platform
- Owner: peterkc
- License: mit
- Created: 2026-04-23T19:16:00.000Z (2 months ago)
- Default Branch: main
- Last Pushed: 2026-06-11T17:28:05.000Z (22 days ago)
- Last Synced: 2026-06-11T19:16:36.519Z (22 days ago)
- Topics: fastapi, fhir, healthcare, home-health, human-in-the-loop, python
- Language: Python
- Size: 470 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 9
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Security: SECURITY.md
- Agents: AGENTS.md
Awesome Lists containing this project
README
# Health Data Platform (HDP)
[](https://github.com/peterkc/health-data-platform/actions/workflows/test.yml)
[](LICENSE)
[](pyproject.toml)
**AI-native workflow and governance layer for healthcare verticals.**
HDP takes clinical documentation from ambient capture through structured
assessment and human-in-the-loop review to EMR write-back, with the consent,
provenance, and audit primitives those workflows demand. Home health is the
first vertical: OASIS-E assessments drafted by an LLM scribe, gated through
confidence-aware review, and synced to EMRs — including the many that expose
no API. It is a reference architecture under active development; no real
patient data is processed anywhere in this repository, and all fixtures are
synthetic.
HDP is **not** a FHIR clinical data repository, and doesn't compete with one.
Pair it with a CDR such as [Medplum](https://github.com/medplum/medplum) or
HAPI: HDP is the workflow that fills the CDR and the governance that controls
what leaves it. The canonical record model treats FHIR R4 as a *projection* —
one output among several (IPS, OMOP) — rather than the storage shape. The
build-vs-pair evaluation is recorded in the
[FHIR surface research](https://github.com/peterkc/health-data-platform/blob/vault/research/fhir-surface.md).
## Architecture
Three layers, dependencies pointing strictly downward:
```
+--------------------------------------------------------------+
| Apps composed deployables (home-health-scribe) |
+--------------------------------------------------------------+
| Verticals domain workflows: OASIS-E assessment, ambient |
| scribe, HITL review, EMR sync |
+--------------------------------------------------------------+
| Platform canonical records | consent | provenance | |
| primitives audit | identity | ingest | agent runtime |
+--------------------------------------------------------------+
FHIR R4 / IPS / OMOP are projections, not storage
```
The platform layer is the governance spine: canonical domain records carrying
trust level and provenance, grant-based consent with 42 CFR Part 2 fail-closed
category semantics, append-only audit, identity linkage, and an agent runtime
whose confidence scores feed the review queue. The vertical layer owns the
domain: OASIS-E modeled as first-class structured data, a
`suggested → reviewed → accepted/edited/rejected` review state machine with a
finalize gate, and a transport-agnostic EMR adapter contract — HTTP APIs where
they exist, browser-level automation where they don't. Schemas and semantics
are specified in the design corpus below; each package README labels its
implementation depth.
## Quickstart
```bash
git clone https://github.com/peterkc/health-data-platform.git
cd health-data-platform
just up # Postgres 18 (pgvector) + Mongo 7; bootstraps .env on first run
just verify # lint + tests (CI mirror)
```
`just dev` (the composed home-health-scribe app) lands with the
[make-it-run epic](https://github.com/peterkc/health-data-platform/issues/12).
## Design research
The `vault` branch carries the design corpus — table designs (30-table DDL
plan, uuidv7 keys), the consent model, FHIR surface evaluation,
interoperability research, and ADRs:
[browse the vault branch](https://github.com/peterkc/health-data-platform/tree/vault).
## License
[MIT](LICENSE)