Flo Health uses Amazon Bedrock to cut medical review time by 60%

Flo Health’s engineering team, in collaboration with AWS, turned a proof of concept into a production AI-powered medical content review and generation system on Amazon Bedrock. The system cut average article review time from seven days to significantly less and tripled content throughput without hiring more medical experts. It addresses the critical bottleneck of medical review, where each article required meticulous fact-checking against a 10-point accuracy checklist. The solution uses a three-layer validation: checking against internal medical guidelines, cross-referencing trusted external sources (clinical decision tools, peer-reviewed journals, regulatory bodies), and then having human experts review AI-annotated content with highlighted rules and source links.

The architecture includes specialized AI Judges for different review dimensions (e.g., factual accuracy, compliance) and a Retrieval Augmented Generation (RAG) system to generate content grounded in reliable references, reducing hallucination risks. The team adapted the MACROS architecture from the PoC to Flo’s content model, focusing on reducing feedback loops between creators and reviewers. Lessons learned include careful prompt engineering, gradual adoption with continuous validation, and designing the system to amplify—not replace—human expertise. This approach allows Flo Health to scale content for millions of users while maintaining medical accuracy and trust.