AI Medical Coding Platform

MedCode Flow turns clinical notes into suggested billing codes, covering ICD-10-CM diagnosis codes, procedure codes and HCPCS codes, and explains each suggestion. For every code it shows a confidence level, the reasoning, and the exact words in the note that support it, highlighted in place. A coder accepts, rejects or edits each suggestion. Once codes are approved, the system builds a draft FHIR R4 Claim, the standard format for the next step in the billing process.

It is built for medical coders and revenue-cycle staff who translate clinical documentation into codes all day, and for coding auditors who need to see why a code was chosen. Organizations that would use it include hospitals, physician practices, and outsourced coding and billing companies that want to speed up coding without giving up accuracy or accountability.

The design keeps the human coder in charge. Suggestions are candidates, never final answers. Every human decision is stored as a permanent record with a full audit trail, and a claim draft can't be created until a person has approved at least one code. The system never submits claims to insurers. Architecturally, it is a Next.js review workspace over a FastAPI service and PostgreSQL. Text extraction and code lookup sit behind clean interfaces, so the local versions can be swapped for AWS services such as Textract, Comprehend Medical and Bedrock, or for a licensed code-set provider, without changing the workflow.

This is the most focused project in the portfolio: an MVP that demonstrates the core review workflow end to end rather than a production platform. Production sign-in, autonomous coding and claim transmission are deliberately out of scope. It also handles code licensing responsibly. CPT procedure codes are copyrighted by the American Medical Association, so the project ships no CPT content and uses an invented demonstration set in its place, with a plug-in point for a licensed source. All data is synthetic.