Skip to content

Builder project · Built at Pulse Foundry AI NYC X ABI Frameworks

WoundScope

Reads messy nursing-home records and flags which wound patients can be billed to Medicare Part B.

WoundScope project preview
Share on LinkedIn

Tip: click @Pulse AI in the post to tag us.

From idea to reality

What we built

## Overview Nursing homes can bill Medicare Part B when they treat a patient's wound — but only if three things are true: the patient has an active wound, active Part B coverage, and the wound is properly documented (length, width, depth + drainage). That information is scattered across insurance records, diagnosis codes, and messy free-text nurse notes, so today a biller checks every patient by hand. It's slow, and billable care gets missed. **WoundScope reads all of it automatically and routes every patient — auto-accept, flag for review, or reject — with the reasoning and the original source text behind each decision.** A biller works only the handful of unclear cases and trusts the rest, because nothing is a black box. ## How it works 1. **Collect** — pulls all records from the PointClickCare API, retrying through its 30% rate-limiting so no patient is ever dropped. 2. **Read** — extracts wound type, location, measurements, and drainage from 4 real note formats and 2 assessment formats (detected by text signature), plus full ICD-10 diagnosis coverage. Handles multiple wounds in one note and repairs garbled text like "Rightlowerle" → "right lower leg". 3. **Decide** — checks the three billing rules and routes each patient, with a plain-English reason and a confidence score. 4. **Show** — an ABI-styled dashboard: a review queue and a per-patient evaluation log showing each rule as pass/fail, the exact evidence behind every value, multi-wound breakdowns, and a billed/dismissed workflow. ## Why it matters - **Recovers revenue** — catches billable wounds (including patients with a second wound a simpler tool would miss). - **Compliance-safe** — never auto-bills on incomplete data; PHI masked by default; every decision audit-logged with provenance (HIPAA-aligned). - **Fast** — 300 patients sorted automatically; the biller reviews ~10, not 300. ## Results (300 synthetic patients) 135 auto-accept · 10 flag for review · 155 reject — zero records dropped, depth recovered for 285/300, every decision explained and evidenced. ## Built as a team We combined the strongest piece each of us built: a resilient ingestion engine, the note/assessment template parsers, full ICD-10 diagnosis coverage + text repair, and a compliance-first dashboard — into one solution. See MERGE.md for who built what. ## Demo notes for judges Open the live app, click any patient for the evaluation log + evidence. Try patient FA-009 to see two-wound detection. Toggle the 🔒 button to see PHI masking. Runs on synthetic data — no real PHI.