Knowledge Graph is a tool that converts any text corpus into a graph of knowledge for use in Graph Augmented Generation or knowledge graph-based question answering.
The tool addresses the problem of extracting meaningful structure from unstructured text by building a semantic network of concepts and their relationships. It works by splitting text into chunks, extracting concepts mentioned within each chunk using a local language model, and inferring relationships between concepts that appear in proximity to one another. Each edge in the resulting graph represents a text chunk containing two connected concepts. The approach prioritizes concepts over named entities, reasoning that concepts like "pleasant weather in Bangalore" create more meaningful graphs than isolated entity names.
The project suits developers who want to build knowledge graphs for retrieval-augmented generation or semantic search without relying on external API calls. It runs entirely on local hardware using Ollama to host an open-source language model, making it economical for personal machines and projects with cost constraints. The tool is presented as a notebook-based implementation, so adoption requires comfort with Jupyter environments and willingness to adapt the provided workflow to specific text sources.
The project shows active development with recent commits and ongoing refinement of the core extraction pipeline. The codebase demonstrates iterative improvement in the graph construction methodology. Documentation includes both conceptual explanation of knowledge graphs and practical implementation details, suggesting attention to helping users understand both the theory and mechanics of the approach.
