Effective troubleshooting agents in enterprise customer support rely on retrieving actionable guidance from similar historical cases. Existing retrieval‑augmented generation (RAG) systems treat support cases as static documents and ignore their multi‑stage, stateful nature. RAFT (Retrieval‑Augmented Framework for Troubleshooting Agents) abstracts each closed historical case into a directed chain of timeline entries and performs retrieval at the entry level, surfacing cases whose intermediate states match the active case and returning the parent‑case trajectory anchored at the matched state; an optional case‑level graph links cases via a configurable similarity representation. We evaluate this retrieval layer directly, avoiding the need for full‑system deployment. Because public multi‑stage troubleshooting data are extremely scarce, we construct a synthetic benchmark from Microsoft Learn Windows Server documentation and complement it with real Apache Jira issues that carry human‑created duplicate labels. RAFT improves Case Hit over vanilla RAG and GraphRAG baselines at every stage of case progress, achieving statistically significant gains over the strongest baseline; the Jira results provide directional evidence that the advantage transfers to real case histories. We release the benchmark, implementation, and the Apache Jira evaluation set.
Review