Building a Backend for ODIN and RUNE: How to Make a Knowledge Extraction Engine

Patrik Kukic's blog post outlines the development of a backend system named BOR, designed to integrate large language models (LLMs) with the Memgraph database to facilitate knowledge extraction from code repositories and notes. Inspired by the concept of using graphs for machine learning from a Ph.D. thesis, the project led to the creation of two applications, ODIN and RUNE, which convert notes and code into knowledge graphs stored in Memgraph. These graphs are queried by specialized agents to provide insights into code structures and note content. The project employs Langchain and OpenAI models for generating Cypher queries from natural language, enabling semantic search using vector embeddings. Additional features include a REST API with FastAPI and bonus functionalities like semantic node-to-sentence associations and code snippet optimization. The blog emphasizes the importance of building an information extraction engine and tools to enable agent reasoning and invites contributions to improve the project.