George He compares local file traversal with pre-indexed retrievalRetrieval is the process of selecting relevant stored information and returning it to an AI system for the current task., explaining why a coding repository's structure does not transfer neatly to a large enterprise document collection. He describes scale, mixed file formats, token costs and permission metadata as reasons to combine keyword and semantic retrievalHybrid search combines retrieval methods, commonly keyword and meaning-based search, so AI applications can handle both exact matches and paraphrased requests. with an agent's ability to inspect specific files.
The talk breaks document search into locating relevant material, traversing metadata, finding exact text and reading rendered content. George He emphasizes structured parsing and page screenshots for tables, diagrams and scanned documents, then addresses synchronization, freshness and storage tradeoffs.
A demonstration uses a prepared collection of Alphabet financial reports to show how hybrid search, grep-like tools and contextual reading can ground a cash-flow answer across several documents. George He presents the harness and its tool primitivesAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction. as the bridge between finding candidate information and producing a traceable resultData provenance records where information came from, how it changed, and which people, systems, or processes handled it..
Watch on YouTube




