Case Study
Equiply Enrichment
Enriches 801 hospital equipment rows with per-vendor serial decoders, resolving 73% locally and calling OpenAI only on the gaps.
Executive Summary
Built for the Equiply Hiring Tournament (NY Tech Week Hackathon), this React dashboard ingests hospital equipment CSVs and fills in manufactured dates and device types with per-vendor serial decoders. OpenAI runs only on deduplicated gap combos to save tokens.
Problem & Constraints
Hospital asset data often carries only manufacturer, model, and serial number. Capital planning committees need manufactured dates, device types, age tiers, and replacement calls.
Architecture
CSV upload → manufacturer normalization → per-vendor serial decoders → static device type map → OpenAI fallback (deduped combos) → capital planning tiers → enriched.csv export.
Methodology
- Built 53 canonical manufacturer|model combos with deterministic date decoders
- Mapped device types across 21 clinical categories
- LLM called only on ~25 unique gap combos (~79% token savings vs row-by-row)
- Added EOL/Review/Active replacement tiers for committee decisions
Results & Metrics
| Metric | Value |
|---|---|
| Total rows | 801 |
| Local decoded dates | ~73.3% |
| Device categories | 21 |
| End of Life assets | 261 |
| LLM combos sent | ~25 (deduped) |
Tech Stack
React, TypeScript, OpenAI API, client-side ETL, CSV export
Future Work
Backend persistence, multi-hospital tenancy, FHIR integration for asset metadata.