Coda

The problem

Medical coding turns what a clinician did into the billable codes a hospital is paid for. It’s slow, error-prone, and unforgiving: under-code and the hospital loses revenue it earned; over-code and it’s fraud. The work demands clinical understanding, current coding rules, and an audit trail for every decision.

What we built

Coda reads clinical operative notes, extracts billable codes with an AI coherence engine, and gives coders and physicians an audited AI copilot to get reimbursement right, defensibly, not by guessing. It runs in live hospital environments, where correctness and privacy aren’t features, they’re the whole job.

How deep it goes

• Real isolation. Multi-tenant by a separate, isolated cloud account per organization. No shared database, no tenant column, actual boundaries between hospitals.
• Audited to the byte. Every AI and human touch of protected health information, read or write, is recorded to a separate, append-only audit log. The AI is held to the same standard as a person.
• Grounded AI. LLM reasoning on AWS Bedrock under a Business Associate Agreement, with retrieval, tool use, and a critic pass, so output can be trusted, explained, and defended.
• Full stack, in production. Real-time streaming chat, background processing, and infrastructure-as-code across the entire system. Designed, built, deployed, and operated by us.

The stack

• AWS (Fargate, RDS, S3, SQS) · infrastructure-as-code
• AWS Bedrock · large language models, retrieval & tool use
• Per-tenant account isolation · least-privilege access
• Append-only audit logging · HIPAA-aligned controls
• Real-time streaming (SSE) · background workers
• Modern typed full-stack web application