Building an editorial AI assistant to support peer review with AWS Generative AI Innovation Center

This article describes how BMJ Group partnered with AWS Generative AI Innovation Center to build an AI-powered editorial assistant for peer review screening using multi-agent architecture.

• AI assistant helps editors screen manuscripts for peer review decisions without replacing editorial judgment
• Eight specialized agents evaluate scope fit, novelty, methodology, ethics, author integrity, and provide synthesis
• Built on Amazon Bedrock AgentCore with Claude models and AWS Step Functions orchestration
• Achieved over 80% success rate across accuracy, error detection, and recommendation alignment
• Methodology agents scored highest in editor helpfulness ratings; scope/ethics agents scored lowest
• Addresses challenge of 50% initial rejection rate with further rejections after peer review
• Structured outputs with source attribution and reasoning build editor trust and transparency
• Key lesson: decompose complex workflows into focused tasks; AI augments expertise, not replaces it
• Future plans include integration with ScholarOne manuscript management system and peer reviewer support

BMJ's approach demonstrates successful AI adoption in scholarly publishing through human-centered design, cross-functional collaboration, and maintaining editorial autonomy in all decisions.