Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
December 09, 2025
·
Paris
Neo4j LangGraph 3D Planning Agent
This talk details building an agentic RAG system using LangGraph and Neo4j to query complex French urban planning regulations and generate 3D feasibility models.
Overview
- Core Problem
- France PLU/PLUi Urban Regulation = huge, fragmented, constantly updated PDFs + annexes.
- Hard to search, slow to analyze, high risk of late non-compliance.
- Data Pipeline
- Automated PDF
- Normalizes zoning rules into a clean structure.
- Everything stored in a Neo4j graph (zones, parcels, articles, files, images).
- Knowledge Graph
- Use LLM to generate Cypher-based retrieval used for Neo4j.
- Full-text + vector search for hybrid rule lookup.
- Each rule answer linked to exact source documents.
- Multi-Agent System (OpenAI + LangGraph)
Agents:
- Regulation Agent (rules + citations)
- Surveyor Agent (parcel + context)
- Model Agent (auto model generation)
- BIM Agent (FAR, height, envelope)
- Supervisor (orchestration + verification)
Workflow triggered by one address + one question.
- 3D & Calculations
- Automatic parcel + building context reconstruction.
Outputs: 3D model, rendered images, feasibility metrics.
- User Interaction
- Chat interface powered by GPT.
Links
Tech stack