Agentic DAG-Orchestrated Planner Framework for Multi-Modal, Multi-Hop Question Answering in Hybrid Data Lakes

ArXi:2603.14229v1 Announce Type: new Enterprises increasingly need natural language (NL) question answering over hybrid data lakes that combine structured tables and unstructured documents. Current deployed solutions, including RAG-based systems, typically rely on brute-force retrieval from each and post-hoc merging. Such approaches are inefficient and leaky, and critically, they lack explicit for multi-hop reasoning, where a query is decomposed into successive steps (hops) that may traverse back and forth between structured and unstructured sources.