🏗️ ConstructGuard AI

AI-Powered Construction Defect Detection & Intelligent Inspection Platform

🔍 Detect • Analyze • Explain • Recommend

ConstructGuard AI leverages Deep Learning, Computer Vision, and Google Gemini 2.5 Flash to automatically identify construction defects from images and generate professional inspection reports with actionable recommendations.

---

🌐 Live Deployment

🚀 Frontend

https://0cb1d45e.constructguard-ai.pages.dev/

⚡ Backend API

https://ayushmsingh2004-constructguard-backend.hf.space/

📚 API Documentation

https://ayushmsingh2004-constructguard-backend.hf.space/docs

---

📖 Overview

ConstructGuard AI is an intelligent construction inspection platform designed to assist engineers, researchers, inspectors, and infrastructure professionals in identifying visible construction defects using Artificial Intelligence.

The platform combines:

• 🧠 Deep Learning Models
• 🤖 Google Gemini 2.5 Flash
• 📷 Computer Vision
• ⚡ FastAPI Backend
• 🌐 Modern React Frontend

to deliver automated defect detection and professional inspection reports.

Unlike traditional image classifiers, ConstructGuard AI first validates whether an uploaded image is genuinely related to construction or infrastructure inspection before performing any defect analysis.

---

✨ Key Features

• 🏗️ Construction Defect Detection
• 🤖 Gemini-Powered Image Validation
• 🧠 Multi-Model Classification Pipeline
• 📊 Severity Assessment
• 📄 AI-Generated Inspection Reports
• 🎯 Specialized Moisture & Structural Models
• 📷 Grad-CAM Visual Explanations
• 📑 PDF Report Export
• 📱 Responsive User Interface
• ☁️ Cloud Deployment

---

🧠 AI Inference Pipeline

Stage 1 – Construction Image Validation

Every uploaded image is first analyzed using Google Gemini 2.5 Flash.

Gemini verifies whether the image contains:

• Buildings
• Walls
• Concrete Structures
• Infrastructure Components
• Construction Sites
• Structural Defects

If the image is unrelated to construction:

❌ Analysis Not Performed

Examples:

Valid:

• Cracks
• Buildings
• Walls
• Concrete Surfaces

Invalid:

• Humans
• Animals
• Vehicles
• Food
• Landscapes
• Random Objects

---

Stage 2 – Router Model

After validation:

Input Image → Router Model

The Router Model predicts:

• Moisture
• Structural
• Surface
• Healthy

---

Stage 3 – Specialist Models

Moisture Specialist

Detects:

• Water Seepage
• Mold
• Algae
• Stains

Structural Specialist

Detects:

• Major Crack
• Minor Crack
• Spalling
• Peeling Paint
• Healthy Surface

---

Stage 4 – Gemini Report Generation

Gemini 2.5 Flash generates:

• Defect Explanation
• Severity Assessment
• Risk Analysis
• Repair Recommendations
• Professional Inspection Summary

---

🏛️ Complete System Architecture

Input Image
     │
     ▼
Google Gemini 2.5 Flash
Construction Validation
     │
     ├────────► Invalid Image
     │              │
     │              ▼
     │      Analysis Not Performed
     │
     ▼
Valid Construction Image
     │
     ▼
Router Model
     │
 ┌───┴────┐
 ▼        ▼
Moisture  Structural
Model     Model
 │          │
 └────┬─────┘
      ▼
Google Gemini 2.5 Flash
Report Generation
      │
      ▼
Final Inspection Report

📂 Dataset

Dataset Repository:

https://drive.google.com/drive/folders/15lIJnX8CfX38zZy8SdtKQtN-jW_hozUF?usp=drive_link

Categories

• Water Seepage
• Mold
• Algae
• Stains
• Major Crack
• Minor Crack
• Spalling
• Peeling Paint
• Healthy Surface

Dataset was curated for construction defect detection and structural inspection research.

---

🧠 Models Used

Model Repository:

https://drive.google.com/drive/folders/1SbuypC_pil5ivAY1XNpoJo8AjghUhqSw?usp=drive_link

Router Model

cg_router.keras

Routes images to specialist models.

Moisture Specialist Model

cg_moisture_specialist.keras

Classifies:

• Water Seepage
• Mold
• Algae
• Stains

Structural Specialist Model

cg_structural_specialist.keras

Classifies:

• Major Crack
• Minor Crack
• Spalling
• Peeling Paint
• Healthy Surface

Frameworks:

• TensorFlow
• Keras
• OpenCV
• NumPy

---

🛠️ Technology Stack

Frontend

• React
• Vite
• React Router
• Framer Motion
• Recharts
• jsPDF

Backend

• FastAPI
• TensorFlow
• Keras
• OpenCV
• Pillow
• NumPy

AI & ML

• Google Gemini 2.5 Flash
• TensorFlow
• Computer Vision
• Grad-CAM

Deployment

• Cloudflare Pages
• Hugging Face Spaces

---

🚀 Local Setup

Clone Repository

git clone https://github.com/AYUSHMSINGH2004/ConstructGuard-AI.git
cd ConstructGuard-AI

Backend

python -m venv venv

Windows:

venv\Scripts\activate

Linux/Mac:

source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Run backend:

uvicorn app:app --reload

Frontend

npm install
npm run dev

---

📡 API Endpoints

Health Check

GET /health

Predict

POST /predict

Form Data:

• file
• api_key

Returns:

• Defect Prediction
• Confidence Scores
• Grad-CAM Visualization
• Gemini Report
• Recommendations

---

📊 Deployment Architecture

graph TD

A[👤 User Uploads Image]

A --> B[🤖 Gemini 2.5 Flash Validation]

B -->|Invalid Image| C[❌ Analysis Not Performed]

B -->|Valid Construction Image| D[🧠 Router Model]

D --> E[💧 Moisture Specialist]

D --> F[🏢 Structural Specialist]

E --> G[🤖 Gemini Report Generation]

F --> G

G --> H[📄 Inspection Report]

H --> I[🌐 Frontend Dashboard]

Loading

🔄 Sequence Diagram

sequenceDiagram

actor User

participant Frontend as Cloudflare Pages Frontend
participant Gemini as Gemini 2.5 Flash
participant Backend as FastAPI Backend
participant Router as Router Model
participant Moisture as Moisture Specialist
participant Structural as Structural Specialist

User->>Frontend: Upload Image

Frontend->>Backend: POST /predict

Backend->>Gemini: Validate Construction Image

alt Invalid Image

Gemini-->>Backend: Not Construction Related

Backend-->>Frontend: Analysis Not Performed

else Valid Image

Gemini-->>Backend: Construction Related

Backend->>Router: Predict Category

alt Moisture

Router-->>Backend: Moisture

Backend->>Moisture: Classification

Moisture-->>Backend: Result

else Structural

Router-->>Backend: Structural

Backend->>Structural: Classification

Structural-->>Backend: Result

end

Backend->>Gemini: Generate Report

Gemini-->>Backend: Recommendations

Backend-->>Frontend: Final Analysis

Frontend-->>User: Display Results

end

Loading

👥 Contributors

👨‍💻 Ayush M Singh

• Project Lead
• AI/ML Development
• Backend Development
• Deployment & Integration
• Testing & Validation

👨‍💻 Venkata Sriram Topalli

• Frontend Development
• Dataset Preparation
• Research & Documentation
• Backend Development
• Testing & Validation

---

📈 Future Enhancements

• Mobile Application
• Multi-User Support
• Cloud Report Storage
• Real-Time Monitoring
• Project Dashboard
• Advanced Analytics
• Multi-Language Support

---

⭐ Support

If you found this project useful:

• Star ⭐ the repository
• Fork 🍴 the project
• Share 📢 with others

---

📜 License

Educational / Academic Use Only