bukosabino/justicio

Embeds BOE documents into a vector database using Spanish-tuned sentence transformers, then retrieves semantically similar articles via approximate nearest neighbor search to provide LLM context. Built on FastAPI, Langchain, and Qdrant, with daily ETL pipelines that chunk documents, generate embeddings, and store metadata for hybrid search capabilities (semantic, keyword, or combined). Supports multiple similarity metrics and operates a free public service while maintaining deployment flexibility for self-hosted instances.

138 stars. No commits in the last 6 months.