RAFT: A Stateful Retrieval-Augmented Framework for Troubleshooting Agents
RAFT retrieves the matching step inside a past support case, then returns the surrounding case trajectory.
The paper frames troubleshooting as a multi-stage process that ordinary RAG flattens into static documents. RAFT turns each closed case into a directed chain of timeline entries and retrieves at the entry level, with an optional graph connecting similar cases. The authors test the retrieval layer on a synthetic Windows Server benchmark and Apache Jira issues with duplicate labels. They report better Case Hit than vanilla RAG and GraphRAG across case stages, with Jira used as directional evidence for real histories. ArXiv · AI/CL/LG's note
The paper frames troubleshooting as a multi-stage process that ordinary RAG flattens into static documents. RAFT turns each closed case into a directed chain of timeline entries and retrieves at the entry level, with an optional graph connecting similar cases. The authors test the retrieval layer on a synthetic Windows Server benchmark and Apache Jira issues with duplicate labels. They report better Case Hit than vanilla RAG and GraphRAG across case stages, with Jira used as directional evidence for real histories. ArXiv · AI/CL/LG's note
score 4