How AI Can Transform Compliance Checking in Architecture
Compliance checking is one of the most essential parts of architectural design, yet it is also one of the most tedious and time-consuming.
Research confirms this reality:
- Around 90% of architecture firms and government bodies still rely entirely on manual compliance checking.
- Professionals spend more than 55 hours per project on this task alone. When managing multiple projects simultaneously, this can easily translate into weeks of work per project.
- Despite the effort, 50–60% of requirements are still missed, leading to costly back-and-forth between designers, consultants, clients, and authorities.
The Foundation
Before developing ARCHAI, we conducted a rigorous Proof of Concept to verify the feasibility of our vision. We focused our study on Begues, a town near Barcelona, where we analysed existing buildings by comparing key data (such as height, density, number of floors, and built area) against the local municipal regulations. The system successfully identified which buildings complied and which did not. Interestingly, the non-compliant cases were primarily older houses built before the current regulations were enacted. This experiment established a solid precedent, proving that our AI-driven approach could accurately interpret complex urban rules.
ARCHAI
This challenge led us to develop Archai, an AI-powered compliance checking plugin designed to integrate directly into the architectural design workflow.
Currently, Archai:
- Works as a Rhino 3D plugin
- Tested with Barcelona’s regulatory framework
- Checks key parameters such as building height, number of floors, and plot area
- Generates clear PDF reports with pass/fail results, regulatory references, and guidance for improvement
ARCHAI Is not meant to replace architects, but to eliminate repetitive manual tasks, reduce errors, and make regulatory constraints clearer from the early stages of design.
Navigating building regulations is often the most tedious part of the design process. ARCHAI changes the game by integrating a powerful Python-driven backend with Rhino 3D to automate compliance checks using Large Language Models (LLMs).
1. The Frontend: Bridging Rhino to Python
The journey begins with a custom Plug-in UI within Rhino. When the user calls ARCHAI, a dedicated window pops up to initiate the scan.
- Layer Scanning: The system automatically scans the Rhino layers to identify design elements.
- Data Packaging: A Python script in the backend processes this spatial data and exports it into a structured EXPORT.JSON file. This file acts as the bridge between the geometry in Rhino and the AI pipeline.
2. Regulatory Digitization with NotebookLM
Before a check can happen, the “regulation” must be readable by the plug-in. ARCHAI uses NotebookLM to turn static PDF regulations into actionable data chunks.
- PDF Injection: The regulation document is uploaded and the text is extracted as a continuous string.
- Normalization: The backend cleans the text, stripping away irrelevant headers, footers, and formatting noise.
- Semantic Chunking: Finally, the text is divided into “Semantic Chunks” paired with metadata for precise referencing later.
3. Geometry as a Graph
To help an AI “understand” a 3D model, ARCHAI converts physical geometry into a Geometry Graph.
- Node Identification: Every 3D object from the Rhino model is injected into the system and transformed into a node.
- Simplification: These nodes are stripped of complex meshes and reduced to simple properties (like dimensions, location, and type), making them lightweight enough for LLM reasoning.
4. The Chroma + RAG Pipeline
This is the “brain” of the operation, where the data becomes searchable memory.
- Tagging & Vectorization: Each regulatory chunk is tagged with logical metadata and converted into numerical embedded vectors.
- Semantic Retrieval: These embeddings are stored in a Chroma database. This allows the system to use RAG (Retrieval-Augmented Generation) to find only the rules that are relevant to the specific geometry being checked.+1
5. LLM Reasoning & Automated Reporting
The final stage is where the “reasoning” happens.
Instant Feedback: The output includes a detailed explanation citing specific rules. This is then compiled into a PDF summary that the architect can open directly within Rhino.
Grounding Context: The system assembles the relevant geometry data and the retrieved regulations into a single prompt.
LLM Analysis: The LLM compares the two, performs logical reasoning, and determines if the design is compliant.
The final result is a seamless interface integrated directly into your existing CAD software. The workflow is designed for maximum efficiency: simply input your project’s location and sync your 3D volumetry. The system processes the local laws and generates a comprehensive report on the compliance of the applicable regulations
Reliability and Accuracy
To ensure that users can fully trust the AI’s conclusions, we implemented a rigorous validation process. We ran the compliance checker multiple times on the same building model to test the consistency of the LLM’s reasoning. Our testing showed that the system delivers the exact same result nearly 92% of the time. This remarkably high success rate demonstrates that ARCHAI isn’t just “guessing”, it provides a stable, reliable, and highly assertive technical audit that architects can depend on for professional projects.
Looking Ahead
Archive is designed to scale and evolve:
- Expansion beyond Barcelona to other cities and countries
- Integration with Revit, AutoCAD, and other BIM platforms
- Potential QGIS integration to include geographic and infrastructural constraints (e.g., utility capacity limits)
- Application beyond regulations to include client requirements and internal design standards
- Internal use within firms to reference past projects and archives, avoiding the need to reinvent solutions
