Inspiration
As software projects grow in complexity, developers struggle to understand large codebases, track changes, and optimize dependencies. Traditional tools lack deep reasoning over repository structures. Inspired by GraphRAG, AI-driven retrieval, and graph analytics, I wanted to create an agentic system that understands, reasons, and visualizes GitHub repositories like never before.
What it does
This app analyzes GitHub repositories by converting them into a knowledge graph, allowing developers to: ✅ Query repository structures in natural language ✅ Visualize dependencies, function calls, and commit histories in 3D ✅ Track design patterns, critical files, and influential contributors ✅ Combine AQL + NetworkX/cuGraph for hybrid graph-based analytics ✅ Prevent context fragmentation & hallucinations in AI-driven code analysis
How I built it
🔹 Data Ingestion: Parsed repositories using Tree-sitter for code structure extraction 🔹 Graph Storage: Used ArangoDB + NetworkX to persist function calls, imports, dependencies 🔹 Query Execution: Implemented LangChain-based AI Agents to process developer queries 🔹 Hybrid Analytics: Integrated AQL for direct queries & cuGraph for GPU-accelerated insights 🔹 Visualization: Used Plotly + Graphviz to create interactive 3D graphs
Challenges I ran into
🚧 Efficient Query Execution: Balancing AQL & NetworkX/cuGraph queries dynamically 🚧 Parsing Large Repositories: Handling commit history with code structures using Tree-sitter 🚧 Contextual AI Responses: Ensuring AI understood repository structures without hallucinations 🚧 Graph Optimization: Structuring millions of relationships efficiently in ArangoDB without losing the node meta or name
Accomplishments that I'm proud of
🏆 Successfully built an agentic system that can reason over repositories 🏆 Developed a hybrid query execution model that combines AQL + GPU graph analytics 🏆 Designed an interactive 3D graph visualization for developers to explore repositories 🏆 Reduced query execution time using optimized ArangoDB graph structures 🏆 Created an AI-powered developer assistant that enhances repository comprehension
What I learned
📌 GraphRAG and other RAG technologies 📌 ArangoDB + NetworkX/cuGraph enables powerful graph-based AI reasoning 📌 Tree-sitter + AI agents can extract meaningful insights from codebases 📌 Hybrid querying (AQL + cuGraph) can drastically improve efficiency 📌 Visualizing repository structures improves developer understanding & debugging
What's next for Next-Gen Agentic App for GitHub Repositories
🚀 Expand AI reasoning for code refactoring suggestions 🚀 VS Code plugin for real-time AI-powered insights 🚀 Integration with CI/CD pipelines for automated risk detection 🚀 Real-time GitHub Webhook support for continuous repository monitoring
Log in or sign up for Devpost to join the conversation.