Multimodal Data Lakes with Chang She, LanceDB

What if the bottleneck in AI isn’t the model... it's retrieval speed? In our latest interview, our Head of AI Bryan Bischof sits down with LanceDB CEO & Co-founder Chang She to unpack why “almost every problem becomes a search problem” because legacy stacks weren’t built for multimodal + agentic workloads 0:00 Riding in a Waymo + quick intros 0:45 What LanceDB is and why search sits under AI 1:49 What “multimodal” really means for enterprise data 3:12 The research-to-production gap (offline vs online) 3:45 Retrieval as the real AI differentiator 5:01 Why LanceDB works for retrieval 6:10 A different vector search architecture (cost + scale) 7:28 Multiple indexes: experimentation and true multimodal search 10:05 One big embedding vs many specialized ones 11:52 Why pgvector/Postgres breaks down at scale 14:10 Prefiltering, permissions, and agents changing search 17:48 The future of vector search in an agent world 21:00 Quick-fire wrap-up