Building a Trustworthy Agentic AI System

Bayer developed the PRINCE platform based on Agentic RAG to efficiently process massive volumes of unstructured preclinical data. It was designed to overcome the limitations of existing keyword-based search, allowing researchers to ask complex questions in natural language and receive accurate answers.

Two core engineering elements were central to building the system:

• Context Engineering: Controls the amount and quality of information the model receives, optimizing how context moves between specialized stages such as Research, Reflection, and Writing.
• Harness Engineering: Focuses on building the scaffolding around the model. This includes orchestration, tool boundary setting, state persistence, retries, fallbacks, validation, reflection loops, observability, and human review processes.

This case study goes beyond simple RAG to cover the technical architecture and engineering decision-making process needed to operate a trustworthy AI Agent system in a real production environment.