"""
Netra AI - Construction Material Classifier
MVP Demo using Claude Vision API
"""
import gradio as gr
import anthropic
import os
from PIL import Image
import base64
from io import BytesIO
# Material classification prompt
CLASSIFICATION_PROMPT = """
You are a construction material classifier for Netra AI. Classify this image into ONE of these 4 classes:
1. **Reet (Sand)**: All grades of sand - from fine powdery sand to coarse gritty sand. Shows smooth to granular texture.
2. **12mm VSI**: Uniform, small, cubical "clean" stones. Very consistent size (~10-15mm). Manufactured aggregate with sharp edges.
3. **Stone**: Large, irregular boulders. Significant size variation, deep voids between rocks. Raw, unprocessed appearance.
4. **GSB (Graded Stone Base)**: Mixed-size material with rocks, gravel, grit AND fine dust/sand filling gaps. Key indicator is fine dust matrix.
Respond in this exact format:
CLASS: [class name]
CONFIDENCE: [High/Medium/Low]
REASONING: [One sentence explanation of visual features that led to this classification]
"""
def classify_material(image):
"""Classify construction material from image using Netra AI"""
if image is None:
return "Please upload an image", "", ""
# Get API key from environment
api_key = os.getenv("ANTHROPIC_API_KEY", "")
if not api_key:
return "❌ System Error", "", "AI service not configured. Please contact support."
try:
# Configure Claude
client = anthropic.Anthropic(api_key=api_key)
# Convert to PIL Image if needed
if not isinstance(image, Image.Image):
image = Image.fromarray(image)
# Convert image to base64
buffered = BytesIO()
image.save(buffered, format="JPEG")
image_data = base64.b64encode(buffered.getvalue()).decode("utf-8")
# Generate classification
response = client.messages.create(
model="claude-opus-4-6",
max_tokens=150,
messages=[
{
"role": "user",
"content": [
{
"type": "image",
"source": {
"type": "base64",
"media_type": "image/jpeg",
"data": image_data,
},
},
{"type": "text", "text": CLASSIFICATION_PROMPT}
],
}
],
)
result_text = response.content[0].text.strip()
# Parse response
lines = result_text.split('\n')
class_name = ""
confidence = ""
reasoning = ""
for line in lines:
if line.startswith("CLASS:"):
class_name = line.replace("CLASS:", "").strip()
elif line.startswith("CONFIDENCE:"):
confidence = line.replace("CONFIDENCE:", "").strip()
elif line.startswith("REASONING:"):
reasoning = line.replace("REASONING:", "").strip()
# Format output
if class_name:
result = f"✅ RESULT:\n\n{class_name}"
else:
result = "⏳ AWAITING CLASSIFICATION"
confidence_display = f"Confidence: {confidence}" if confidence else ""
reasoning_display = f"💡 **Analysis:** {reasoning}" if reasoning else result_text
return result, confidence_display, reasoning_display
except Exception as e:
error_msg = str(e)
if "API_KEY_INVALID" in error_msg or "401" in error_msg:
return "❌ Invalid API Key", "", "Please check your Gemini API key"
elif "quota" in error_msg.lower():
return "❌ API Quota Exceeded", "", "Your API key has exceeded its quota"
else:
return f"❌ Error: {error_msg[:100]}", "", "Please try again"
# Custom CSS
custom_css = """
#header {
text-align: center;
background: linear-gradient(90deg, #667eea 0%, #764ba2 100%);
padding: 20px;
border-radius: 10px;
color: white;
}
#result {
font-size: 72px;
font-weight: 900;
text-align: center;
padding: 50px 30px;
border-radius: 20px;
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
color: white;
margin-bottom: 25px;
box-shadow: 0 8px 16px rgba(0,0,0,0.2);
text-transform: uppercase;
letter-spacing: 2px;
line-height: 1.2;
}
#confidence {
font-size: 24px;
font-weight: 600;
text-align: center;
padding: 15px;
background: #f0f9ff;
border-radius: 10px;
margin-bottom: 20px;
}
#reasoning {
font-size: 14px;
text-align: left;
padding: 15px;
background: #f9fafb;
border-radius: 8px;
margin-bottom: 20px;
color: #4b5563;
}
#about {
font-size: 13px;
color: #6b7280;
padding-top: 20px;
border-top: 1px solid #e5e7eb;
}
"""
# Build Gradio interface
with gr.Blocks(css=custom_css, title="Netra AI - Material Classifier") as demo:
gr.Markdown(
"""
""",
elem_id="header"
)
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### 📸 Upload Material Image")
image_input = gr.Image(
type="pil",
label="Construction Material Photo",
height=400
)
classify_btn = gr.Button("🚀 Classify Material", variant="primary", size="lg")
gr.Markdown(
"""
**Supported Materials:**
- 🏖️ Reet (Sand)
- 🔷 12mm VSI (Uniform Aggregate)
- 🪨 Stone (Large Boulders)
- 🏗️ GSB (Mixed Graded Base)
"""
)
gr.Markdown("### 📷 Try Sample Images")
gr.Examples(
examples=[
["sanity_test_dataset/12mm_VSI_01.jpeg"],
["sanity_test_dataset/Reet_01_fine.jpeg"],
["sanity_test_dataset/Stone_01.jpeg"],
["sanity_test_dataset/GSB_01.jpeg"],
],
inputs=image_input,
label=""
)
with gr.Column(scale=1):
result_output = gr.Markdown("", elem_id="result")
confidence_output = gr.Markdown("", elem_id="confidence")
reasoning_output = gr.Markdown("", elem_id="reasoning")
gr.Markdown(
"""
### 📊 About This Demo
This is a proof-of-concept for **Netra AI's** automated material classification system.
**Technology:**
- Vision AI for real-time material identification
- Prevents grade fraud at construction sites
- Automates dispatch logging
**Use Cases:**
- Gate monitoring at crusher plants
- Dispatch verification
- Quality control
---
**Netra AI** - *Making every tonne count.*
For commercial inquiries: [Contact Us](mailto:guptaraghu321@gmail.com)
"""
)
# Connect button
classify_btn.click(
fn=classify_material,
inputs=[image_input],
outputs=[result_output, confidence_output, reasoning_output]
)
# Launch
if __name__ == "__main__":
demo.launch(
server_name="0.0.0.0",
server_port=7860,
share=True
)