AIRA_2: Overcoming Bottlenecks in AI Research Agents

AIRA_2 addresses three structural bottlenecks in AI research agents: synchronous single-GPU execution limiting throughput, validation-driven generalization gaps, and fixed single-turn LLM operator limitations. The system introduces asynchronous multi-GPU worker pools, a Hidden Consistent Evaluation protocol, and ReAct agents that dynamically debug—achieving 71.8% percentile rank on MLE-bench-30, improving over the prior best of 69.9%.