For anyone who has ever turned to the internet during a health scare, the stakes of medical content accuracy are deeply personal. Flo Health, the company behind the world's leading women's health app, produces thousands of medical articles each year and is now partnering with AWS to use generative AI to verify that content remains current and accurate at scale.
The collaboration, detailed in a recent AWS blog post, centers on a custom-built system called MACROS (Medical Automated Content Review and Revision Optimization Solution). Developed through a partnership with the AWS Generative AI Innovation Center, MACROS is designed to process large volumes of medical content against credible scientific sources, flag potential inaccuracies or outdated information, and propose updates based on the latest research and guidelines.
Medical knowledge evolves constantly. Treatment protocols change, new studies emerge, and clinical guidelines are updated, sometimes in ways that directly contradict previous recommendations. For a platform like Flo Health, which serves millions of users worldwide with information on women's health topics, keeping pace with that evolution through manual review alone is an enormous challenge. Each article review can take hours, and the sheer volume of content makes it difficult for even a large team of medical experts to stay ahead.
MACROS addresses this by automating the initial layers of content review. The system, powered by Amazon Bedrock, breaks articles into sections, reviews each against specified medical rules and guidelines, and identifies paragraphs that may no longer align with current standards. Only flagged sections move to a revision stage, where the AI suggests updates while maintaining the original tone and style. The revised content is then reassembled into a complete document for human expert review.
The architecture runs on a suite of AWS services, including Amazon Elastic Container Service, AWS Step Functions, AWS Lambda, Amazon Textract, and Amazon S3. A Streamlit-based user interface allows Flo Health's medical experts to interact directly with the review results, view content statistics, and make manual adjustments, reinforcing that the system is built to augment, not replace, human expertise.
During the proof-of-concept phase, MACROS achieved over 90% recall in identifying content requiring updates while maintaining 80% accuracy. The system also applied medical guidelines more consistently than manual reviews and significantly reduced the time burden on medical experts. Processing speed exceeded initial targets, achieving a 10x faster review completion than manual processes.
One of the more notable features is the Rule Optimizer, which extracts actionable guidelines from unstructured documents such as PDFs of medical research. It categorizes rules by type, medical condition guidelines, treatment-specific guidelines, and changes to health advice, and assigns priority levels to guide the review process.
The system also uses a tiered approach to model selection. Simpler tasks, such as text chunking, use smaller models, while complex reasoning tasks, such as content review and revision, leverage larger models, optimizing both performance and cost.
Flo Health and AWS plan to continue developing the solution, with Part 2 of the series expected to cover production scaling challenges. For now, the proof of concept demonstrates a practical path to maintaining the integrity of medical content at a scale that manual processes alone cannot sustain.
The company, which has built a robust security and identity infrastructure to protect user data, continues to invest in technology that supports its mission of delivering trustworthy health information.