The Fifth Elephant 2025 Annual Conference CfP

Abstract:

Preclinical R&D is full of buried insights—locked in silos, old PDFs, and structured databases. What if a single system could surface the right answer or write a regulatory report in minutes? At Bayer, we co-developed PRINCE, a GenAI-powered research assistant that helps scientists navigate over 2 million+ data chunks and complex structured data, bringing order, speed, and precision to R&D workflows.

This talk unpacks how we built a production-grade, domain-specific chatbot that unifies unstructured and structured data across OpenSearch, Athena, and Snowflake, powered by a multi-agent system capable of handling everything from complex queries and doing deep research to drafting compliance-critical regulatory documents.

We’ll walk through:

• Architecting an advanced RAG pipeline with query expansion, cross-encoders, hybrid search and metadata filters.
• Architecting a robust text2sql feature with semantic layer so that natural language queries can be answered over complex structured data.
• A multi-agent system that accesses both unstructured and structured data, capable of solving complex questions and writing long, compliance-critical reports with high accuracy.
• Strategies to ensure factuality, traceability, and compliance—because GenAI in pharma isn’t just about speed, it’s about trust.
• Real-world impact: how the system cuts document creation time from weeks to days, and is shaping next-gen decision-making in drug development.

Whether you’re in healthcare, legal, or enterprise search, you’ll walk away with key patterns for orchestrating GenAI systems in real-world enterprise settings—with understanding of why explainability, observability, and user trust must be first-class citizens in GenAI systems.