Multi-agent systems

How multiple LLM/VLM agents collaborate, debate, verify, and refine each other's outputs. Research on emergent communication protocols, consensus-building, and multi-agent self-play for improving reasoning and task completion quality.

Building scalable frameworks for multi-agent pipelines: task routing, tool use, memory systems, and self-correction loops. How to design agent architectures that are reliable, composable, and can scale from single tasks to complex workflows.

Training and prompting agents for distinct roles (critic, coder, researcher, planner) and studying how role assignment affects team performance. Research on when specialization outperforms generalist agents and how to dynamically allocate roles.

Using adversarial debate and cross-agent verification to improve output quality. Research on how agents can catch each other's mistakes, reduce hallucination through mutual critique, and produce more reliable final outputs.

Systems that generate their own training signal through synthetic data, self-reflection, and iterative refinement. Investigating feedback loops where agents evaluate their own outputs, generate preference pairs, and continuously improve without human annotation.

Developing evaluation frameworks for multi-agent systems: measuring coordination efficiency, task decomposition quality, communication overhead, and emergent capabilities that arise from agent interaction at scale.