The workflow starts from high-level protocol requirements that pull data from sources like clinicaltrials.gov, PubMed, and regulatory guidelines, which are converted into prompts for LLM-based protocol writers. It then generates multiple protocol drafts with consistent structure but varying content, scores them on key quality dimensions with configurable evaluators, and routes passing versions to data synthesizers that merge protocol details with real-world data (for example CMS, CDC, SEER) to produce privacy-preserving synthetic cohorts aligned to each protocol.