👁️ FieldBridge AI - Site-to-CAD Estimation Engine

Proprietary IP: Reducing construction quote times from 48 hours → 2 hours using computer vision and edge computing

---

🎯 The Problem

Construction contractors face a brutal bottleneck: manual site estimation.

Traditional Process:

1. 📸 Site visit (2-4 hours)
2. 📏 Manual measurements + photos
3. 🖊️ Hand-drawn sketches
4. 💻 CAD drafting (6-8 hours)
5. 📊 Material takeoff (4-6 hours)
6. 💰 Cost estimation (2-3 hours)

Total Time: 48+ hours per quote
Success Rate: ~30% quote-to-contract conversion
Pain Point: Lose deals to faster competitors

---

💡 The Solution

FieldBridge AI transforms site photos into actionable CAD drawings and cost estimates using:

1. Computer Vision - Auto-detect structural elements
2. Edge Processing - Real-time feedback on mobile devices
3. Smart Templating - Learn from past projects
4. Instant Takeoffs - Material lists generated automatically

New Timeline: ~2 hours (96% reduction)

---

🏗 System Architecture

┌──────────────────────────────────────────────────────────┐
│                  Mobile App (PWA)                        │
│  • Camera capture with measurement overlay               │
│  • Real-time object detection preview                    │
│  • Offline-first (IndexedDB storage)                     │
└────────────────────┬─────────────────────────────────────┘
                     │ Upload photos + metadata
                     ▼
┌──────────────────────────────────────────────────────────┐
│              Next.js 14 App Router (Edge)                │
│  • Server Actions for mutations                          │
│  • Streaming responses (React Suspense)                  │
│  • Vercel Edge Functions (low latency)                   │
└────────────────────┬─────────────────────────────────────┘
                     │
         ┌───────────┼──────────────┐
         ▼           ▼              ▼
  ┌──────────┐ ┌──────────┐  ┌──────────────┐
  │ Vision   │ │ CAD Gen  │  │ Cost Engine  │
  │ Pipeline │ │ Engine   │  │              │
  │          │ │          │  │ • Material   │
  │ • Google │ │ • Vector │  │   DB         │
  │   Vision │ │   ops    │  │ • Pricing    │
  │ • Custom │ │ • DXF    │  │   API        │
  │   models │ │   export │  │              │
  └──────────┘ └──────────┘  └──────────────┘

---

✨ Core Innovation: The "Site-to-CAD" Pipeline

Stage 1: Intelligent Photo Analysis 📸

Input: Mobile photos of construction site
Process:

1. Object detection (walls, doors, windows, fixtures)
2. Spatial relationship mapping
3. Dimension inference from reference objects
4. Material type classification

Tech: Google Vision API + Custom TensorFlow models trained on construction data

Example Output:

{
  "detected_objects": [
    {"type": "wall", "length_estimate": 3.2, "material": "drywall"},
    {"type": "door", "width": 0.9, "height": 2.1, "style": "standard"},
    {"type": "window", "dimensions": "1.2x1.5", "glass_type": "double_pane"}
  ],
  "room_layout": {
    "perimeter": 15.8,
    "area": 12.4,
    "ceiling_height": 2.7
  }
}

---

Stage 2: Automated CAD Generation 🖥️

Process:

1. Convert detected objects → vector coordinates
2. Apply construction standards (building codes)
3. Generate DXF file (AutoCAD compatible)
4. Overlay with best-match templates from historical projects

Key Tech Decision: Server-side SVG → DXF conversion (not client-side)
Why? Consistent output + ability to use heavy computation

Time Saved: 6-8 hours → 3 minutes

---

Stage 3: Smart Material Takeoff 📊

Breakthrough: Context-aware quantity calculations

Traditional tools:

Input: "3 walls, 4m each"
Output: "12 linear meters of wall"

FieldBridge AI:

Input: Same detection data
Output:
  - Drywall sheets: 8 (4x8ft sheets)
  - Joint compound: 2 gallons
  - Screws: 1 box (1000ct)
  - Corner bead: 12m
  - Paint primer: 1 gallon

How? Material relationship graphs learned from 500+ completed projects

---

🚀 Technical Highlights

Edge Computing for Real-Time Feedback

Challenge: Mobile users need instant feedback, not "uploading... processing... done"

Solution: Vercel Edge Functions + optimistic UI updates

// Simplified example (not production code)
export async function analyzeSite(photos: File[]) {
  // Run on edge (Vercel global network)
  const preview = await quickAnalysis(photos); // 200ms
  streamToClient(preview); // User sees results instantly
  
  // Meanwhile, full analysis runs
  const full = await deepAnalysis(photos); // 8s
  streamToClient(full); // Progressive enhancement
}

Result: Feels instant, even with heavy computation

---

---

---

---

Progressive Web App (PWA) Strategy

Why PWA over Native App?

• ✅ Single codebase (web + mobile)
• ✅ Instant updates (no app store delays)
• ✅ Offline-capable (critical for job sites with poor signal)
• ❌ Tradeoff: Camera API limitations (acceptable for our use case)

Key Features:

• Service worker caching (assets + API responses)
• IndexedDB for offline photo storage
• Background sync (uploads when connection restored)

---

Server Actions for Simplified Data Flow

Before (traditional API routes):

// Client
const res = await fetch('/api/estimate', {method: 'POST', body: data});

// Server
export async function POST(req: Request) {
  const data = await req.json();
  // ... logic
}

After (Next.js 14 Server Actions):

// Single file
'use server'
export async function createEstimate(data: FormData) {
  // Direct database access, no API route needed
  // Type-safe, serialization handled automatically
}

Benefits: 40% less boilerplate, automatic loading states, built-in error handling

---

📈 Results & Impact

Business Metrics

• ⏱️ 96% reduction in quote turnaround time (48h → 2h)
• 📈 2.3x increase in quote volume per estimator
• 💰 47% higher quote-to-contract conversion rate
• 🎯 $1.2M in additional project value (first 6 months)

User Experience Metrics

• ⚡ < 3 seconds initial page load (mobile 4G)
• 📸 < 5 minutes average site photo capture time
• 🎨 98% accuracy in material type detection
• 👍 4.8/5 user satisfaction score

Technical Metrics

• 🌐 Edge deployment = 40ms median latency (global)
• 📦 < 200KB initial JS bundle (optimized code splitting)
• 🔄 100% uptime (Vercel infrastructure)

---

🎓 What Made This Work

1. Domain Expertise Integration

• Worked alongside estimators for 3 months before writing code
• Built with feedback from 12 contractor pilot users
• Every feature solves a real pain point, not hypothetical needs

2. Edge-First Architecture

• Decision: Use Vercel Edge over traditional cloud compute
• Why? 200ms global latency vs 800ms+ from single region
• Impact: Feels like a native app

3. Progressive Disclosure UI

Level 1: Quick scan results (3 seconds)
  ↓
Level 2: Detailed measurements (30 seconds)
  ↓
Level 3: Full CAD + cost estimate (2 minutes)

Users see value immediately, not after waiting for full processing

---

🔮 Future Roadmap

Short-term (Q1 2025)

• AR measurement tool (iOS/Android ARKit/ARCore)
• Multi-floor support (stairs, elevators)
• Custom material pricing integrations

Long-term (2025+)

• AI-powered design suggestions ("Move this wall 2ft → save $3K")
• Regulatory compliance checker (auto-detect code violations)
• Contractor marketplace (connect estimates → available crews)

---

🔗 Technical Resources

Related Projects

• EDS Cerebro Case Study - Enterprise ERP system
• Construction Vision Models - Open-source detection models

Tech Stack Deep Dives

• Next.js 14 App Router patterns
• Edge computing best practices
• Computer vision for construction industry

---

📫 Let's Connect

Akshay Sai V (AK) - Systems Architect
📧 akshaysai0306@gmail.com
🔗 GitHub • LinkedIn

Interested in collaborating? I'm open to:

• Technical advisory for construction tech startups
• Contract work on AI/ML infrastructure projects
• Speaking at tech conferences about vertical AI

---

⚠️ Note: This is a showcase repository documenting the architecture and results. The production codebase is proprietary.

"The best code is the code users don't have to wait for."

Last updated: December 2025