LLM-Codex-Reference-Vault

This project bridges the gap between static Markdown/PDF technical files and LLM interactions by parsing documents into connected Chunk, CodeBlock, and Concept nodes within Neo4j, completely eliminating hallucinations via the Model Context Protocol (MCP).

Features GPU-Accelerated PDF Processing: Offloads heavy PDF-to-Markdown conversion to a remote Docling instance. Image Stripping & Sanitization: Retroactively cleans extraneous Base64 images from Markdown files to preserve LLM token limits. Semantic Neo4j Graph: Constructs a connected graph mapping abstract technical concepts down to their raw code implementations. Automated Dev.to Integrations: Ships with ingest_devto.py for headless API ingestion of highly-rated articles via cron jobs. Native MCP Integration: Plug-and-play compatibility with Neo4j's official Model Context Protocol server.