Bayer and Thoughtworks describe PRINCE, an agentic RAG system for preclinical research

Bayer AG and Thoughtworks describe PRINCE, a cloud-hosted platform that evolved from keyword search to an intelligent research assistant for preclinical drug development. PRINCE integrates decades of safety study reports and structured metadata using Agentic Retrieval-Augmented Generation (RAG) and Text-to-SQL.

The authors introduce two engineering lenses: context engineering, which governs what information each specialized agent receives and how context moves between research, reflection, and writing agents; and harness engineering, which provides orchestration, tool boundaries, state persistence, retries, fallbacks, validation, reflection loops, observability, and human-in-the-loop review around the models to maintain control and reliability.

The system addresses data silos, limited search capabilities, and time-consuming manual analysis in Bayer’s preclinical research environment by consolidating siloed structured metadata and unlocking unstructured PDF study reports that remain the authoritative source despite incomplete metadata.

PRINCE’s agentic RAG pipeline includes agents for clarifying user intent, planning and reflection, research, data validation and sufficiency, and answer synthesis and formatting, with built-in transparency, explainability, and governance to support regulatory compliance.