Overview
Jira Graph RAG is a learning-focused project built to understand GraphRAG end-to-end. It ingests a public Jira dataset, normalizes issues and link relationships, then combines FAISS-based semantic search with Neo4j graph expansion to produce hybrid-ranked answers with traceable citations. The system demonstrates how semantic retrieval alone misses relational context, and how graph neighbor expansion improves coverage on relation-heavy queries.
The Motivation
I built this project to deeply learn GraphRAG architectures by implementing the full pipeline from scratch. The goal was to understand hybrid retrieval tradeoffs, evaluate evidence quality with real metrics, and build practical debugging workflows for RAG systems.
Key Features
End-to-end GraphRAG pipeline from raw Jira data to citation-first answers
Hybrid retrieval fusing FAISS semantic seeds with Neo4j graph neighbors
Evaluation framework with Recall@10 and MRR metrics
Streamlit app with Answer, Evidence, and Graph Debug views
Configurable embedding backends (local TF-IDF or Gemini)
Technologies
Tags
GraphRAGLearning ProjectJiraRAGKnowledge GraphEvaluation
Type
Personal Project
