BuildPulse

Inspiration

Construction delays cost the industry $1.6 trillion globally every year. We discovered that 98% of projects face delays, yet 52% of contractors still track progress with paper clipboards and highlighters. Enterprise solutions cost $50,000+, pricing out small contractors who need help the most. We built BuildPulse to democratize AI-powered construction tracking—making it accessible to everyone from solo contractors to major developers.

What it does

BuildPulse is an AI-powered construction progress tracking platform that:

Analyzes construction photos using computer vision to automatically detect completion percentage
Supports 3D building models (GLB/GLTF) with interactive zone assignment for mesh-level tracking
Predicts project delays using machine learning on progress velocity data
Calculates financial impact showing exact dollar costs of delays ($15K+/day)
Generates AI video reports for stakeholders with zone-by-zone breakdowns
Identifies what's complete vs missing (e.g., "Framing done, roofing still missing")

How we built it

Frontend

Next.js 16 with App Router and React 19 for a modern, performant UI
React Three Fiber + Three.js for interactive 3D model visualization with orbit controls
Tailwind CSS 4 for styling
Radix UI components for accessible, polished interactions
Recharts for progress tracking visualizations

Backend

MongoDB + Mongoose for scalable project and zone data storage
Auth0 for secure authentication
MongoDB GridFS saved the project by helping us store 3D models

AI/ML

Google Gemini 2.0 Flash for construction photo analysis — detects building components (roof, walls, foundation, etc.), estimates completion percentage, and lists what's done vs. missing
Gemini AI also powers our smart zone naming suggestions when users click on 3D model meshes
Linear regression for progress velocity predictions and estimated completion dates
OpenNote SDK for generating AI-narrated video reports for stakeholders

Challenges we ran into

3D mesh selection: Getting click-to-select working on GLB models required building a custom SelectionManager with raycasting and material-based grouping
AI accuracy: Tuning Gemini prompts to reliably identify construction stages (framing vs. finishing) took extensive iteration
Real-time sync: Keeping 3D viewer state, database, and UI in sync when analyzing photos required careful state management
File size limits: 3D models exceeded Supabase's limits, so we implemented base64 encoding with MongoDB storage for the hackathon

Accomplishments that we're proud of

✅ Full working prototype with 3D model support, AI analysis, and video report generation
✅ Smart component detection: Gemini correctly identifies building elements like "Roof", "Foundation", "Exterior Wall" from mesh geometry
✅ Financial impact calculator that shows exact delay costs based on project budget
✅ Scroll animations and premium UI with glassmorphism, 3D progress cylinders, and interactive cards
✅ End-to-end flow: From project creation → 3D upload → zone assignment → photo check-in → AI analysis → video report

What we learned

Gemini's vision capabilities are incredibly powerful for domain-specific analysis when given proper context
Three.js in React requires careful memory management and camera positioning for user-friendly 3D experiences
Construction industry pain points are massive—there's genuine need for accessible AI tools
Hackathon scope management: Starting with core features (AI analysis) before polish (animations) was crucial

What's next for BuildPulse

📱 Mobile app for on-site photo capture with offline support
📊 Multi-project dashboards for construction firms managing portfolios
🚁 Drone footage analysis for large-scale site monitoring
🤝 Contractor marketplace connecting delayed projects with available workers
🌱 Carbon footprint tracking to quantify sustainability impact of early delay detection