Back to work
AI/ML
Permit & Zoning RAG Assistant
Government / civic
A retrieval-augmented assistant that answers building-permit and zoning questions with citations, drawing on official municipal code, permit PDFs, and zoning tables.
The Challenge
Permit and zoning rules are scattered across dense code, PDFs, and tables — making preliminary guidance slow and hard to source reliably.
What We Built
- An ingestion pipeline for mixed-format sources (HTML/XML code, PDFs, structured tables) into a searchable corpus.
- Semantic search with pgvector embeddings; citation-first answers that explicitly avoid unsupported claims (no guaranteed approvals).
- An optional OCR fallback (Azure AI Document Intelligence) for scanned inputs.
Highlights
- Citation-first answers over official municipal sources
- Mixed-format ingestion (code, PDFs, tables)
- Semantic search with pgvector
- Guardrailed scope — guidance, not guarantees