REVERE: Reflective Evolving Research Engineer for Scientific Workflows
This addresses the challenge of reproducing results in heterogeneous research environments, offering a novel method for continual learning in AI agents, though it is domain-specific to scientific workflows.
The paper tackled the problem of poor generalization in prompt-optimization techniques for research-coding workflows by introducing REVERE, a framework that learns from global context and reusable heuristics, resulting in performance improvements of 4.50% on SUPER, 3.51% on ResearchCodeBench, and 4.89% on ScienceAgentBench over prior state-of-the-art.
Existing prompt-optimization techniques rely on local signals to update behavior, often neglecting broader and recurring patterns across tasks, leading to poor generalization; they further rely on full-prompt rewrites or unstructured merges, resulting in knowledge loss. These limitations are magnified in research-coding workflows, which involve heterogeneous repositories, underspecified environments, and weak feedback, where reproducing results from public codebases is an established evaluation regime. We introduce Reflective Evolving Research Engineer (REVERE), a framework that continuously learns from Global Training Context, recognizes recurring failure modes in cross-repository execution trajectories, distills them into reusable heuristics, and performs targeted edits across three configurable fields: the system prompt, a task-prompt template, and a cumulative cheatsheet. REVERE, via this reflective optimization framework, improves performance over prior state-of-the-art expert-crafted instructions on research coding tasks by 4.50% on SUPER, 3.51% on ResearchCodeBench, and 4.89% on ScienceAgentBench across their respective metrics. These results demonstrate that agents equipped with mechanisms for continual learning and global memory consolidation can meaningfully evolve their capabilities over time.