Can AI solve the biggest problem in drug discovery? Time!
summary
Bayer and Thoughtworks build an AI‑powered Preclinical Information Center using Generative AI and Retrieval Augmented Generation (RAG) to turn 50 years of reports (≈100 GB, ~2 million chunks) into a production Q&A and report writer. Architecture combines keyword filtering, weighted hybrid semantic search, cross‑encoder reranking and multi‑agent orchestration so complex queries break into subtasks. Results move simple tasks from days to minutes, complex queries from weeks to minutes, and IND drafting from months to minutes while flagging citations and facing PDF, observability and hallucination challenges. search, ask, do
AI-written summary · about
chapters (13)
- 0:00Drug discovery challenges
- 3:14Preclinical testing costs
- 7:28Gen AI for research questions
- 9:38RAG for information retrieval
- 13:54Semantic search scaling
- 18:34Gen AI in research demos
- 21:50Agentic systems for queries
- 24:23Writing assistant for IND reports
- 27:16Trustworthy Gen AI challenges
- 31:13Evaluation and production readiness
- 32:36PDF extraction hurdles
- 33:42Grounded answers for compliance
- 34:03Evaluation speed optimization
related talks
30mAI in Drug Discovery
36mBuilding Agentic AI Assistants: Lessons from productionising agents in Preclinical R&D
23m"Deliver AI at Lightspeed: Streamlined approach to AI strategy"
44mRAG for a Medical Company: Technical and Product Challenges AI & Data
45mGenerative AI in medicine: Improving patient & provider experiences
23mAI in the Life Science Sector
43mAdvanced Retrieval-Augmented Generation Techniques
08mEngineering the Future of Pharma: AI-Driven Drug Discovery