Automatically extract, structure, and semantically search through the full UK Approved Documents — from PDF to vector embeddings in seconds.
From PDF parsing to semantic search, the pipeline handles the heavy lifting so you can focus on building.
Split master PDF into individual pages, extract text with layout preservation, and generate structured metadata.
Vector embeddings enable natural-language queries across all approved documents. Find exactly what you need.
LLM-powered section classification, cross-reference detection, and compliance mapping for each clause.
Production-grade database with vector similarity search, full-text indexing, and relational integrity.
RESTful API for querying processed data, triggering re-processing, and integrating with other tools.
Automatically detect and extract diagrams, figures, and tables from approved documents with SVG export.
Six stages from raw PDF to searchable knowledge base.
Download and validate the merged Approved Documents PDF with SHA-256 integrity check.
metadata.jsonSplit the master PDF into 1,541 individual page files for parallel processing.
pdf_pages/Extract sections, diagrams, appendices, and cross-references from each page.
page_extractorStore structured data in PostgreSQL with full relational integrity.
seeder.pyGenerate embeddings for all sections using an OpenAI-compatible LLM endpoint.
vector_db_managerQuery processed data via the FastAPI backend or the web interface.
localhost:8000Clone the repo, run ./setup.sh dev, and you're live in minutes.