# AI for Architecture ## Computer vision for building intelligence The Building Safety Act 2022 changed everything. Manual compliance checking cannot keep up with the golden thread requirements for higher-risk buildings. AI can. ## The Regulatory Shift ### The Building Safety Act changed everything The Building Safety Act 2022 introduced the most significant reform to building safety regulations in a generation. For higher-risk buildings, the requirements are substantial. The Act applies to residential buildings with seven or more storeys, or those exceeding 18 metres in height. It introduces the concept of the “golden thread” - comprehensive, continuously updated documentation that traces every material change to a building design from initial concept through operation and maintenance. Maintaining this golden thread manually is not viable at scale. The administrative burden of tracking every design change, flagging implications for fire safety, structural stability, and accessibility, and generating updated compliance documentation requires automation. This is where computer vision and AI deliver transformative value - not as a replacement for professional judgement, but as a tool that surfaces the information architects and safety teams need, faster and more reliably than manual processes. #### Golden Thread Requirement Continuous audit log of every design change, with auto-flagged implications for regulated aspects of building safety. #### Higher-Risk Building Regime Enhanced oversight for residential buildings of 7+ storeys or 18m+ height, with mandatory gateway approvals at design, construction, and completion stages. #### Building Safety Regulator New regulator within the HSE responsible for overseeing the safety and performance of all buildings, with specific powers for higher-risk buildings. #### Duty Holder Obligations Clear accountability chain from client through principal designer and contractor, with ongoing responsibilities throughout the building lifecycle. #### BIM Integration Digital building models increasingly expected as part of the golden thread, enabling automated compliance checking against thousands of code requirements. ## The Transformation ### Manual process vs. AI-powered analysis Every stage of design review and compliance checking is fundamentally changed when computer vision is applied to architectural documents. #### Compliance Review Before Hours of manual checking against building codes. Error-prone, slow, and expensive. With AI Automated scanning in seconds. AI flags non-compliance issues before they become costly rework. Hours to seconds #### Floorplan Analysis Before Architects manually measure, annotate, and cross-reference layouts against regulations. With AI Computer vision extracts room boundaries, door positions, and measurements automatically. 95%+ accuracy #### Golden Thread Documentation Before Fragmented records across teams. No single source of truth for design decisions. With AI Continuous audit log with auto-flagged implications for fire safety, structural stability, and accessibility. Full traceability #### Design Iteration Before Compliance feedback arrives late in the process, forcing expensive redesigns. With AI Real-time compliance checking as designs evolve. Issues detected and flagged immediately. Shift-left validation ## Technology Stack ### Purpose-built for architectural analysis General-purpose AI models are not production-ready for architecture, engineering, and construction tasks. We build domain-specific pipelines that combine multiple models for reliable, accurate results. **YOLO Object Detection** Real-time detection of architectural elements - doors, windows, walls, stairs, fire exits, and accessibility features - across floorplan images and BIM exports. **SAM Segmentation** Segment Anything Model for precise boundary detection of rooms, zones, and spatial regions. Pixel-level accuracy for measurement and area calculations. **CLIP Classification** Vision-language model for classifying architectural elements by function and context. Connects visual features to regulatory requirements using natural language. **OpenCV Image Processing** Core image processing pipelines for boundary detection, measurement extraction, perspective correction, and coordinate system alignment. **PyTorch Model Training** Custom training workflows for continuous accuracy improvement. Domain-specific fine-tuning on architectural datasets to exceed general-purpose model performance. **FastAPI API Layer** High-performance API serving predictions at scale. Async processing for batch analysis of large document sets with real-time progress tracking. ## Processing Pipeline ### From floorplan to compliance report A typical document goes through multiple analysis stages, each adding a layer of intelligence to the final output. 01 ### Ingest PDF, DWG, IFC, or image upload. Format detection and preprocessing. 02 ### Detect YOLO identifies architectural elements. Doors, walls, stairs, exits. 03 ### Segment SAM extracts room boundaries. Pixel-level precision for measurements. 04 ### Classify CLIP maps elements to regulatory categories. Context-aware labelling. 05 ### Report Compliance findings, measurements, and flagged issues in structured output. ## Case Study ### Arcus: AI-powered building intelligence Arcus needed a system that could analyse floorplans for compliance and design errors faster and more reliably than manual review. Architects and safety teams were spending hours on checks that should take minutes. We built an AI platform combining YOLO for object detection, SAM for segmentation, and CLIP for classification, with OpenCV pipelines handling boundary detection and measurement extraction. Custom training workflows ensure the system gets more accurate over time with domain-specific data. Proof that AI, when trained and tuned properly, can transform an industry. ### Review time **Before** Hours per document **After** Seconds per document ### Compliance confidence **Before** Dependent on reviewer experience **After** Consistent, measurable accuracy ### Design decisions **Before** Feedback late in the process **After** Safer, smarter decisions earlier in projects ## Build smarter. Comply faster. Whether you are an architecture practice, a construction firm, or a proptech startup, we can help you apply computer vision and AI to your compliance and design review workflows.