Dr-DCI: Scaling Direct Corpus Interaction via Dynamic Workspace Expansion
Researchers have developed Dr-DCI, a new framework for agentic search over large corpora that dynamically expands a local workspace. This approach combines the scalability of retriever-mediated interfaces with the precision of Direct Corpus Interaction (DCI) operations. By pulling relevant documents into an evolving workspace, Dr-DCI aims to improve efficiency and performance across various scales, outperforming existing methods on benchmarks like Browsecomp-Plus and Wiki-18 QA. AI
IMPACT Enhances agentic search capabilities by improving efficiency and scalability for interacting with large document corpora.