jiangnanboy/llm-knowledge-graph

The project is a straightforward implementation of a common RAG-adjacent workflow: using LLMs for named entity recognition and relationship extraction to build a knowledge graph. With only 26 stars and 3 forks over a year, it lacks the momentum or community of major competitors like LangChain (GraphIndex) or LlamaIndex. The technical approach is a standard pipeline of OCR/PDF-parsing followed by prompting, which is increasingly becoming a native capability of frontier models (e.g., GPT-4o, Gemini 1.5 Pro, and Chinese-specific models like Qwen-VL or DeepSeek). These models handle multimodal inputs directly, making the manual OCR-to-prompting pipeline in this repo redundant. The project serves more as a personal tutorial or experiment rather than a defensible piece of infrastructure. There is no unique dataset, fine-tuned model, or specialized algorithm that provides a moat against either frontier labs or more mature open-source frameworks.