DaprTS - Deterministic NLP Pipeline to CDS
Deterministic, auditable NLP for transforming clinical notes into CDS-ready artifacts.
Why I Built This
Most clinical NLP systems feel like black boxes, which is risky when outputs influence care decisions or policy checks. I built DaprTS to turn unstructured notes into CDS-ready artifacts with deterministic steps that can be inspected and audited. The motivation was to make NLP outputs trustworthy enough for real clinical workflows, not just demos.
What I Implemented
- A visible planner loop so every transformation step can be inspected.
- Terminology grounding across SNOMED CT, LOINC, RxNorm, ICD-10-CM, and UCUM.
- Strict
$validate-codechecks before accepting mapped concepts. - Structured outputs in JSON and CSV for downstream CDS use.
Stack
FastAPI + Streamlit with configurable terminology endpoints through config/terminology.yaml.
Outcome
A practical NLP-to-CDS bridge that is built for reviewability, not just one-off demos.