---
AI Summary: "\\- Developed an AI-powered Medicare coding analyzer prototype in under\
  \ 4 weeks.  \n- Integrated with existing SMART on FHIR EHR using TypeScript and\
  \ Angular.  \n- Automated CMS code lookup, EHR data processing, and LLM prompts\
  \ via Azure functions.  \n- Delivered a functional UX, supporting multiple LLM backends\
  \ and Medicare mappings."
Anonymous: false
Assignee:
- ''
Last Edited Time: '2026-01-25T22:17:00+00:00'
case_study_510ks: []
client_logo:
- 18abd5b7-a754-80b4-a0fa-def68cdfa073
client_name: Untitled
date: '2025-11-09'
featured: false
medical_panel: Digital Health
name: Built an AI-Powered Medicare Coding Analyzer in 4 Weeks
services:
- 9dc4f55a-b5cb-419f-8bae-5e11d6b11fee
summary: Delivered a working AI prototype that automates Medicare coding analysis
  in under 4 weeks.
tags:
- Web-App
- AI/ML
testimonials:
- 38241a44-bb71-4b52-ac39-fd821e12f430
---

## The Problem

Medicare uses risk-adjusted models for reimbursement, with a hierarchical system of condition categories (CMS-HCC), diagnoses (DRG, ICD-10), treatments, and codes - which are not always easy to follow and assign on a case-by-case basis. Identifying situations where a code has been missed or could be substituted for a closer match is a win-win for both the patient and the care provider; the care provider gets a higher reimbursement value, and the patient is more likely to receive full services.

A client was looking to apply a semi-automated process to the (currently) manual task of reviewing a patient case file and determining what medicare coding should be applied, and came to us for building out a prototype.

## Highlights

We quickly iterated on an MVP, built on top of their existing SMART on FHIR EHR application, and delivered a functional prototype that met the requirements. The integration used a combination of deterministic mapping tables, existing patient demographics, conditions, and medications (automatically pulled via FHIR), and engineered LLM prompts with injected context. All of this was wrapped up in an easy-to-use UX, intended to be used and understood by multiple healthcare professionals.

- Prototype showed successful results, even with limited patient records pulled from the EHR
- Prototype could generate an automated human-readable text report with the medical coding analysis, to be saved back to the EHR
- Final prototype met (and exceeded) requirements, while also having gone through rounds of UX iteration
- Included support for multiple LLM backends (Gemini, OpenAI GPT)
- Included support for multiple versions of Medicare mapping tables
- Extracted LLM analysis code to a scalable Azure function, built within the existing project infrastructure pipeline
- Automated a demo environment spin-up, with a containerized EHR sandbox

## What We Did

- Built our prototype as a drop-in module within an existing Angular SMART-on-FHIR monorepo - using TypeScript, Angular templating, CSS, etc.
- Created an automated script to take raw CMS code lookup tables from Medicare (provided as spreadsheets) and extract and transform values into a queryable local relational data store
- Created an automated process to combine EHR data with an engineered LLM prompt and deterministic mapping lookups, implemented as an Azure function
- Iterated on UX with client team on a weekly basis

## Timeline

- Functional UX was achieved within weeks, allowing for fast iteration with client on changes
- Entire MVP was completed in under 4 weeks of billable time
- Met with client on a weekly basis