Automated Clinical Metadata Harmonization Dashboard

cBioPortal hosts 400+ cancer genomics studies whose clinical metadata is wildly heterogeneous — treatment fields alone have 24+ variants. The mentor's team built MetaHarmonizer, a 4-stage ML pipeline using sentence-transformer semantic matching and ontology resolution (NCIT, MONDO, UBERON), but it has no web interface, no audit trail, and no cBioPortal-ready export. This project wraps the pipeline in a production-ready curator dashboard: upload a raw metadata CSV, review and correct automated column mappings with confidence scores and top-5 alternatives, validate ontology term assignments, monitor quality KPIs, and export a cBioPortal-compatible TSV with a full JSON audit report. I have already built a working full-stack prototype (React + FastAPI) integrating the real SchemaMapEngine — processing 141 columns in 2.6 seconds — with 20+ REST endpoints, an on-demand LLM rematch endpoint, and an automated evaluation pipeline on the provided eval set. Deliverables: (1) production-grade backend with PostgreSQL persistence and role-based auth, (2) polished React dashboard with all needed curator pages, (3) formal ML benchmarking report, (4) CI/CD pipeline and (5) full documentation and cBioPortal deployment guide.