Скачать или смотреть Enhancing Automated Code Generation with GraphRAG - Project Showcase

Unlike traditional Retrieval Augmented Generation (RAG) methods that use vector embeddings, my project leverages Graph-based Retrieval Augmented Generation (GraphRAG) to effectively capture dependencies, deprecations, limitations, version control, and best practices within API documentation.

By representing APIs as graphs, I trained Claude 3.5 Sonnet to generate accurate code for an unfamiliar package, achieving a 90% success rate across 50 diverse requests. This scalable solution is ideal for API-first companies, enabling them to effortlessly set up and manage GraphRAG representations of their APIs and SDKs, thereby optimizing automated code generation with large language models and streamlining development workflows.

Speaker - Joaquin Coromina

Tech stack - Supabase, Vercel, Neo4j, Google Cloud Run Functions