Building Reliable Agentic AI Systems
7 hours ago
- #Retrieval-Augmented Generation
- #Agentic AI Systems
- #Pharmaceutical Research
- PRINCE is a cloud-hosted platform developed by Bayer AG with Thoughtworks to address pharmaceutical challenges in drug development by integrating decades of safety study reports.
- It evolves from keyword-based search to an intelligent research assistant using Agentic Retrieval-Augmented Generation (RAG) and Text-to-SQL for complex queries and drafting regulatory documents.
- Key engineering focuses on context engineering (shaping and routing information between specialized agents) and harness engineering (orchestration, recovery, observability) for reliability and control.
- The system prioritizes trust through transparency, explainability, human-in-the-loop integration, and features like citations for fact verification.
- Architecture includes a LangGraph-based orchestration, multi-agent workflow (Researcher, Writer, Reflection Agents), and data retrieval from OpenSearch and Amazon Athena.
- Resilience is ensured via error handling, state persistence, retries, LLM fallbacks, and monitoring with tools like Langfuse and CloudWatch.
- Iterative development emphasizes early user feedback, accuracy over cost optimization, and continuous improvement based on real-world usage.