From Self-Evolving Synthetic Data to Verifiable-Reward RL: Post-Training Multi-turn Interactive Tool-Using Agents

Researchers developed EigenData, a framework combining self-evolving synthetic data generation with reinforcement learning to train AI agents for multi-turn tool usage and dialogue. The system achieved 73% success on Airline tasks and 98.3% on Telecom benchmarks, matching frontier models while eliminating the need for expensive human annotation.

• →EigenData framework combines synthetic data generation with verifier-based reinforcement learning for training tool-using AI agents.
• →The system achieved 73.0% pass rate on Airline tasks and 98.3% on Telecom tasks, matching frontier model performance.
• →Self-evolving data synthesis eliminates the need for expensive human annotation in training complex AI behaviors.
• →The approach uses hierarchical multi-agent architecture with executable checkers to improve data quality and reliability.
• →Results demonstrate a scalable pathway for bootstrapping complex tool-using behaviors in AI systems.