Scaling Medical Content Review at Flo Health with Amazon Bedrock

Flo Health has developed an AI-driven system to streamline the review and creation of medical content, leveraging Amazon Bedrock. The system evolved from a proof of concept at the AWS Generative AI Innovation Center into a scalable, production-ready solution. It employs specialized AI Judges for different review dimensions such as medical accuracy, legal compliance, and brand style, each tailored with custom prompts and examples to enhance precision.

The architecture adapts the MACROS pattern, splitting content into manageable sections for targeted evaluation. This modular approach allows for independent improvement of each Judge, maintaining high standards of accuracy and safety. The system integrates retrieval-augmented generation (RAG) to incorporate external medical sources, ensuring content is grounded in reliable information.

The AI content generation pipeline retrieves relevant guidelines and knowledge base materials from Amazon S3, structures content with Claude models via Amazon Bedrock, and validates outputs through multiple checks. This process reduces manual effort, accelerates content creation, and maintains strict compliance with medical standards.

By capturing and reusing corrections and feedback, Flo's system continuously improves, reducing routine corrections by 80% and review time by 60%. The approach emphasizes augmenting human expertise rather than replacing it, with AI assisting at pre-screening, drafting, and review stages. This methodology enables Flo to scale content production without expanding its medical team, ensuring high-quality, accurate health information for millions of users.

The implementation highlights the potential of domain-specific AI models and structured feedback loops in sensitive fields like healthcare. It also demonstrates how cloud-based generative AI services like Amazon Bedrock can be tailored for high-stakes applications, balancing innovation with safety and reliability.