Accelerating SDTM Engine Development and Adoption with a Multi-Agent AI Framework and RAG-Based Assistant

About this session

Part of Session 7, Track A: Foundational Standards - SDTM (Chair: Alana St. Clair, CDISC).

Derivation logic for CDISC SDTM is increasingly delivered as reusable functions within standardization engines. While this modular strategy enhances consistency, authoring high-quality functions remains a significant investment due to variability in clinical data structures, conventions, therapeutic areas, and protocols.
This paper introduces a human-in-the-loop, multi-agent AI workflow that accelerates SDTM Engine function development and validation. Specialized agents handle requirements interpretation, code generation, test-case design, and documentation, each under developer-led review. We demonstrate the approach using functions covering reference-data lookups, conditional transformations, and data-integrity validation, verified through regression suites spanning functional, negative, and edge-case scenarios.
To support adoption, a complementary RAG-based assistant provides context-aware guidance on function selection, configuration, and troubleshooting from a Git-based knowledgebase.
A tiered governance model master-prompt constraints, developer review, and independent validation ensures AI accelerates delivery without compromising quality or traceability. Initial results show meaningful reductions in iteration cycles and improved throughput.

Speakers

• PD

  Proma Debnath

  Merck & Co.

  Speaker

• SM

  Srinivas Malipeddi

  Merck & Co.

  Speaker