Bayer PRINCE

Bayer PRINCE (Preclinical Information Center) is Bayer's production preclinical-research assistant described in the Martin Fowler case study "Building Reliable Agentic AI Systems." It evolved from Search over structured preclinical metadata, to Ask with RAG over unstructured study PDFs, to Do with an agentic workflow that can orchestrate complex research tasks and support regulatory-document drafting.^{source: martin-fowler-bayer-reliable-agentic-ai-systems-2026.md}

The platform is a concrete enterprise example of agentic-rag: a LangGraph workflow coordinates clarifying user intent, thinking and planning, retrieval from structured and unstructured sources, reflection on evidence sufficiency, and final answer synthesis. Its data layer combines OpenSearch for vectorized reports, Athena for structured data, PostgreSQL for LangGraph checkpointed agent state, DynamoDB for application state, and internal LLM platforms with model fallbacks.^{source: martin-fowler-bayer-reliable-agentic-ai-systems-2026.md}

PRINCE is especially useful as a reference case for production-llm-reliability because its reliability mechanisms are explicit: persisted workflow state, retries at LLM-call and node level, user-initiated retry from the failed node, LLM-provider fallback, transparent intermediate steps, granular citations to underlying documents, Langfuse traces, RAGAS evaluations, and daily live-traffic evaluation.^{source: martin-fowler-bayer-reliable-agentic-ai-systems-2026.md}

Resources

Building Reliable Agentic AI Systems