medical-coder

Inspiration

Slow healthcare system. High insurance costs. Overworked medical staff.

What it does

Automates and standardizes medical coding.

How we built it

Strands. FloTorch AI Gateway. A lot of headaches.

Challenges we ran into

Data collection. Medical code data is hard to find and is formatted differently across organizations. A lot of medical code information is also proprietary. Determinism. LLM models provide a lot of creativity that I had to tone down for repeatable behavior while keeping reasoning capability.

Accomplishments that we're proud of

Robust datasets. Vector databases. Clean agentic graph flow. Amazon Nova is able to quickly use my custom medical code database search tool to retrieve accurate medical codes based on diagnoses and procedures in the prompt.

What we learned

AI agents are very capable in data processing. Strands allows rapid development and offers robust capabilities. Managing agent state is a must.

What's next for medical-coder

Larger datasets. Price calculation based on coverage policies. Patient profile capabilities.

Built With

• bedrock
• flotorch
• nova
• opensearch
• python
• strands