Search papers, labs, and topics across Lattice.
This paper introduces HERA, a hierarchical framework for multi-agent RAG that adaptively evolves both the orchestration of agents and their individual prompts. HERA uses reward-guided sampling to optimize query-specific agent topologies and role-aware prompt evolution to refine agent behaviors based on credit assignment. Experiments on knowledge-intensive benchmarks show HERA achieves a 38.69% improvement over baselines, demonstrating efficient coordination and robust reasoning through emergent self-organization.
Multi-agent RAG gets a 38% performance boost by letting agents dynamically rewire themselves and evolve their own prompts based on experience.
Multi-agent Retrieval-Augmented Generation (RAG), wherein each agent takes on a specific role, supports hard queries that require multiple steps and sources, or complex reasoning. Existing approaches, however, rely on static agent behaviors and fixed orchestration strategies, leading to brittle performance on diverse, multi-hop tasks. We identify two key limitations: the lack of continuously adaptive orchestration mechanisms and the absence of behavior-level learning for individual agents. To this end, we propose HERA, a hierarchical framework that jointly evolves multi-agent orchestration and role-specific agent prompts. At the global level, HERA optimizes query-specific agent topologies through reward-guided sampling and experience accumulation. At the local level, Role-Aware Prompt Evolution refines agent behaviors via credit assignment and dual-axes adaptation along operational and behavioral principles, enabling targeted, role-conditioned improvements. On six knowledge-intensive benchmarks, HERA achieves an average improvement of 38.69\% over recent baselines while maintaining robust generalization and token efficiency. Topological analyses reveal emergent self-organization, where sparse exploration yields compact, high-utility multi-agent networks, demonstrating both efficient coordination and robust reasoning.
