The Auth-to-Claim Engine (ATCM) project is a reconciliation demonstration that addresses the following business challenge:
"A health plan processes millions of claims. Every day, a subset cannot be automatically adjudicated because authorization and claim records don't align. These exceptions create operational cost, payment delays, provider abrasion, and revenue leakage." (ChatGPT)
My goal was to build a credible application that was, largely, based on my work on a medical pharmacy modernization initiative in 2024.
The screenshots shown here are a result of this early prototyping exercise using Cursor. The entire demonstration took several hours on a single afternoon, but it was based on a week-long engineering exploration I conducted in late 2025, the first post of which is available here.
While I am satisfied with this initial demonstration, there is still much more that can be done, including database integration (including the use of real source data with PHI exceptions), more robust CI integrations, and the ability to employ additional agents to address exception cases (vs. the manual processes showcased here, even with the use of GenAI). This latter point is especially valid given the sheer volume of claims and authorizations that result in exceptions daily.
You can access the demonstration here. I'm currently working on version 2.
Technical profile: Python FastAPI + Pandas backend, React/Vite/TypeScript/Tailwind dashboard, CSV-backed data, OpenAI for explanations, and Vercel for deployment
Copyright © Adrian Daniels, 2004-2026