New approach replaces passive document retrieval with active navigation, allowing AI agents to explore hierarchical knowledge structures and combine evidence across multiple sources

Corpus2Skill system converts document collections into navigable skill directories, enabling LLM agents to see corpus organization rather than passively consuming search results

Uses iterative clustering and LLM-generated summaries to build a tree-structured hierarchy that agents can drill into with progressively finer detail levels

Agents gain ability to reason about search paths, backtrack from unproductive routes, and combine evidence scattered across different knowledge branches

Tested on WixQA enterprise dataset, addressing limitations of traditional RAG systems that lack visibility into corpus structure