Skip to content

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

ClaimReady

We find the billable cases before the biller finds a headache.

Share on LinkedIn

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

From idea to reality

What we built

Overview A data pipeline that ingests patient records from a PointClickCare API, extracts wound documentation from clinical notes and assessments, and produces a routing decision for each patient. Results are displayed in a biller-facing dashboard showing who to bill, who needs review, and who to skip. Problem Wound care billing under Medicare Part B requires active MCB coverage, a documented wound type, measurements, and drainage level. This triage currently happens manually. The goal was to automate data collection and routing across 300 patients in three facilities, handling messy free-text notes, rate-limited APIs, and ambiguous clinical documentation. Approach Ingestion -- fetch all patients across facilities 101, 102, 103. Check coverage first and reject non-MCB patients immediately, avoiding unnecessary API calls for ineligible patients. For MCB patients, fetch diagnoses, notes, and assessments concurrently using a queue with 10 workers and retry logic with exponential backoff and jitter. Extraction -- parse wound fields from both structured assessments and free-text notes. Assessments are processed first as they are more reliable. Regex handles common formats like 2.9 x 2.8 cm, stage references, and drainage keywords. Notes fill in any gaps the assessment missed. Routing -- three deterministic decisions. Auto accept when all required fields are present. Flag for review when the wound is documented but a field is missing, for example a pressure ulcer without a stage. Reject when the patient is not MCB or no wound can be identified. Dashboard -- filterable table with charts, sorting, pagination, and a per-patient detail modal showing extracted fields and a plain-English reason for the routing decision. Notes All 10 flagged patients had pressure ulcers with measurements and drainage but no documented stage, giving the biller a clear and specific next action. The 32 MCB rejects had measurement data in their notes but no identifiable wound type, so the pipeline correctly refused to guess rather than risk a bad claim. Routing logic is fully deterministic so every decision is auditable. The LLM fallback is optional and never makes the final routing call directly.