Autonomous Multi-Agent Scientific Research: A 361-Project Case Study in Thermoelectric Materials Discovery

Keywords: autonomous multi-agent systems, thermoelectric materials, scientific discovery automation, materials informatics

TL;DR: A hierarchical multi-agent system autonomously executed 361 thermoelectric materials research projects, demonstrating that AI agent coordination can maintain scientific rigor while achieving unprecedented research throughput.

Abstract: We present an autonomous multi-agent research system that conducted large-scale scientific discovery without human intervention. The system executed 361 thermoelectric materials projects across multiple research cycles, demonstrating unprecedented scale in autonomous research. Our hierarchical architecture spans experimental validation, theoretical physics checks, ML modeling, and documentation synthesis. Key achievements include: (1) 100\% physical constraint compliance ($0 \le zT \le 3.2$) across all projects; (2) knowledge accumulation through 7,500 RAG entries; (3) resilience validated through real-world agent malfunction recovery. While complete metrics were not systematically captured, distributed data extraction revealed performance documentation in 59.8\% of projects. This work establishes autonomous multi-agent coordination as a viable paradigm for accelerating scientific discovery.

Submission Number: 76