The Unstructured Data Problem No One Solved (Until Now)

Context Window Podcast is back for its 2026 season! Ed Anuff and Anant Jhingran are kicking off the first podcast of the new year with a topic that’s quietly blocking most AI success stories: Unstructured data. Not models. Not agents. We’re joined by Peter Staar, Principal Research Scientist, AI for Knowledge at IBM, to talk about: 👉 Docling, Langflow, and what it actually takes to make unstructured data usable for AI. If you’ve ever built a RAG pipeline that looked right but behaved… unhinged, assumed OCR + chunking = “good enough” or wondered why AI pilots stall after the demo phase: This one’s for you! What we dig into: 👉 Why unstructured data isn’t a side quest, it’s the problem 👉 What most teams miss when they treat documents as “just text” 👉 How Docling changes ingestion by saving structure (layout, hierarchy, semantics) 👉 How Langflow fits into real, visual ingestion-to-agent workflows 👉 Why ingestion quality now determines AI ROI Peter also shares why 2026 will be the year of applications built on Docling.