Skip to content

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

Pulse.wce

Turning messy clinical notes into auditable Medicare Part B billing decisions.

Pulse.wce project preview
Share on LinkedIn

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

From idea to reality

What we built

## What we built **Pulse** is a Medicare Part B wound care eligibility engine. It ingests 300 synthetic patients from a rate-limited mock PCC API, extracts wound data from four note formats (SOAP, prose, multi-wound, and unstructured Envive narratives), and routes each patient to `auto_accept`, `flag_for_review`, or `reject` — with every decision traced back to the exact field that drove it. ## How it works - **Sync → Extract → Decide** pipeline, fully idempotent (survives 429s and partial syncs without losing or duplicating data) - Every wound field gets a status, not just a value: `known`, `unknown_missing`, or `unknown_unparseable`. - **Rules engine is the source of truth.** A decision tree (84% CV accuracy) runs alongside it as an advisory signal only — it predicts what the rules would say, but never overrides a billing decision - An Unknown Risk Score (green/yellow/red) tells billers not just *whether* a patient is eligible, but *how confident* we are in the data behind that call. - A FastAPI dashboard with live sync, AI-generated patient summaries (Baseten GLM, constrained to never invent codes or numbers), and a biller override workflow with logged justifications ## Why it matters Most teams will show "eligible / not eligible." Real billing data isn't that clean — out of 300 patients, only 15 were clean enough to auto-accept. Pulse is built around that reality: it surfaces exactly what's missing or conflicting instead of hiding it, keeps every routing decision auditable back to a deterministic rule, and uses ML to assist.