Repository Issue Insight Assistant
Client
Mid-size enterprise
Sector
IT Services
Revenue
€ 59M (2024)
Client
Mid-size enterprise
Sector
IT Services
Revenue
€ 59M (2024)
From challenge...
Enable fast, reliable answers about open issues, recent changes, and authorship across repositories. Reduce time spent searching and correlating information; improve triage efficiency and speed up issue resolution.
...to result
35%less time spent searching and 10% faster triage and assignment!
From challenge...
Enable fast, reliable answers about open issues, recent changes, and authorship across repositories. Reduce time spent searching and correlating information; improve triage efficiency and speed up issue resolution.
....to result
35%less time spent searching and 10% faster triage and assignment!
The solution
RAG assistant that lets developers ask natural language questions about open issues, recent changes, ownership and links across many repositories. It combines vector search with metadata filters and a graph database service to surface relationships and returns answers with evidence and tables.
- Cross repo issue Q and A
- Graph enriched tables for owners and links
- On premise deployment with traceable evidence
Instead of manually browsing multiple sources, developers ask questions and get a single grounded answer across repositories, including open items, recent changes, authors, links and status. The assistant supports this flow by combining retrieval signals and returning responses enriched with tables that summarize what matters. Through conversation, users can:
- Check what is open and what changed recently across repositories
- Identify authors and owners faster
- See linked items surfaced as relationships and returned in structured tables
Technology stack
The solution is built around a hybrid retrieval layer that combines semantic search with relationship enrichment to maximize precision and context completeness.
- Vector DB semantic search over issues, comments and related artifacts using an open source embedding model, with metadata filtering and boosting
- Structured metadata stored per chunk, including repository, labels, status, authors, assignees, timestamps, components and links
- Graph DB enrichment via an external graph service to retrieve relationships between issue, pull request, commit, author and repository
- RAG orchestration that merges retrieved passages and graph facts, then generates answers with an on premise LLM under 30B parameters
- Application stack with React front end, Node.js middleware and chat history persisted in PostgreSQL via LangChain