MedCode

🩺 Inspiration

While working at Vanderbilt Medical Center and Williamson County Hospital, and later interning at Kaiser Permanente, I saw how medical coders work long hours reading clinical notes and trying to assign the right ICD-10 codes. Even small mistakes could delay insurance claims or cost the hospital thousands.

Across the U.S., hospitals spend over \$30 billion each year on manual billing and coding. On average, coders spend 10–15 minutes per chart, and up to 80% of medical bills contain some form of coding error.

This inspired us to ask: Can AI help coders code faster and more accurately — without needing to change their workflow?

💡 What it does

MedCode is an AI-powered assistant that reads clinical documents (like discharge summaries or doctor’s notes) and suggests ICD-10 diagnosis and procedure codes. It helps reduce the time it takes to review a patient case and ensures the codes are accurate, consistent, and explainable.

Here's how it helps:

• Cuts manual coding time from 15 minutes to under 1 minute per chart.
• Suggests the most likely ICD-10 codes and explains the reasoning.
• Helps hospitals avoid costly claim denials and improve billing compliance.

🛠 How we built it

We used:

• Claude 3 Sonnet via Amazon Bedrock, accessed through LiteLLM, as our core model to read and interpret the documents.
• Temporal to manage the flow — like reading, clarifying, and finalizing codes.
• The system is built to be modular and simple — coders can upload a document, get suggestions, and download results.

🏆 Accomplishments that we're proud of

• We built a working agent that understands real medical documents and returns accurate ICD-10 codes with reasoning.
• It can scale to hundreds of charts, saving hours of manual work for coding teams.
• It’s simple to use — no training needed, just upload and review.

🚀 What's next for MedCode

• Add support for procedure codes (CPT/HCPCS).
• Add a feedback loop so users can approve or correct codes — and improve the model over time.
• Expand to support multi-language coding and ICD-11 in the future.

Built With

• ai
• bedrock
• temporal