Agentic Problem Solving from Jira
LLM agent with tool-calling for automated code investigation from Jira tickets
Built a Langchain-based agentic system at Cummins that automatically investigates and proposes solutions for Jira tickets.
Architecture:
- Retrieves stored procedures, table definitions, and documentation from SQL databases
- Embeds codebase documentation with incremental on-load updates using ChromaDB
- Tool-calling agent orchestrator with capabilities to read source files, query databases, and parse Jira tickets
- Manually annotated evaluation dataset from past Jira tickets for benchmarking retrieval accuracy
Tech: Python, Langchain, ChromaDB, Jira API