The latest release of neo4j-labs/llm-graph-builder is v0.8.7. This repository is a web application that processes unstructured data such as PDF, DOC, TXT, YouTube videos and web pages with Large Language Models (LLMs) and LangChain, builds a knowledge graph by extracting entities and relationships, and stores it in Neo4j.
The center of this repository is bundling material from several sources into a single flow and then adding conversation to the graph. Local files, as well as GCS and S3, and material from web addresses, can all be fed into the same pipeline, and you choose the LLM and schema yourself. A knowledge graph means a structure that connects entities and relationships pulled out of documents, and the resulting graph can be viewed in Neo4j Bloom. There is also a feature to look up, as metadata, which source a given response came from while you converse with the data. The license is Apache-2.0, the primary language is Jupyter Notebook, GitHub stars are 5283, and the creation date is 2024-01-11.
That said, the collected information contains no changes for v0.8.7. Only the release number can be confirmed, and there is no way to know what this version changed. No performance figures or benchmarks are presented either, so in the 10-point scale evaluation only the momentum item is left at 1.0, while the freshness, technical and quality items are empty. The design of keeping source metadata is useful when you trace a response back, but it does not substitute for accuracy or processing speed. Until the changes are confirmed in the release notes, it is hard to judge what this version improved.