From Threads to Trajectories: Multi-LLM Pipeline Extracts Community Knowledge from GitHub Issue Discussions

arXiv·medium signal

Tackles the problem that complex GitHub issue threads contain valuable resolution knowledge buried in unstructured, fragmented, multi-participant discussions. Proposes a multi-LLM pipeline that transforms issue discussion threads into structured knowledge trajectories — capturing not just the final resolution but the diagnostic reasoning path. Directly useful for teams building developer-facing AI tools: the extracted trajectories can serve as training data for issue-triage agents or as searchable knowledge bases for recurring problems.