Aurinara

Preparing SDTM deliverables is slow and error-prone because clinical source data uses inconsistent structures, names, and formats. Programmers must manually interpret metadata, map variables to SDTM domains, define joins and derivations, apply controlled terminology, write programs, and maintain traceability across multiple documents. Built for the Meta AI Developer Hackathon, Clinical AI Artifact Workbench uses Meta’s OpenAI-compatible AI API, specialized agents, and MCP integrations with clinical standards sources to automate much of this workflow. It generates reviewable SDTM mappings, R or SAS programs, Define-XML, and submission documentation through staged human-review gates, combining generative AI with deterministic validation and artifact generation.