README.md

---
title: FDAM AI Pipeline
emoji: "\U0001F525"
colorFrom: red
colorTo: yellow
sdk: gradio
sdk_version: "6.3.0"
app_file: app.py
pinned: false
suggested_hardware: l4x4
---

# FDAM AI Pipeline

**Fire Damage Assessment Methodology v4.0.1** - An AI-powered system that generates professional Cleaning Specifications / Scope of Work documents for fire damage restoration.

## Features

- **AI-Powered Image Analysis**: Uses Qwen3-VL vision model to detect fire damage zones, conditions, and materials
- **FDAM Compliant**: Implements Fire Damage Assessment Methodology v4.0.1 standards
- **Automated Calculations**: Air filtration, sample density, labor estimates per FDAM formulas
- **Professional PDF Output**: Generates ready-to-use Scope of Work documents
- **Session Persistence**: Save and resume assessments via browser localStorage

## How to Use

1. **Project Info**: Enter project details, facility classification, and assessor information
2. **Building/Rooms**: Add rooms with dimensions (length, width, ceiling height)
3. **Images**: Upload fire damage photos and associate with rooms
4. **Observations**: Record qualitative observations (odor, soot, char, etc.)
5. **Generate**: Click "Generate Assessment" to run AI analysis and produce documents

## Technical Details

### Model Stack (~38-43GB VRAM)
- **Vision**: Qwen3-VL-30B-A3B-Thinking-FP8 (~30-35GB) - Reasoning-enhanced analysis with structured JSON output
- **Embeddings**: Qwen3-VL-Embedding-2B (~4GB)
- **Reranker**: Qwen3-VL-Reranker-2B (~4GB)

### Zone Classifications
- **Burn Zone**: Direct fire involvement, structural damage
- **Near-Field**: Adjacent to burn zone, heavy smoke/heat exposure
- **Far-Field**: Smoke migration only, light deposits

### Condition Levels
- **Background**: No visible contamination
- **Light**: Faint discoloration, minimal deposits
- **Moderate**: Visible film/deposits
- **Heavy**: Thick deposits, surface texture obscured
- **Structural Damage**: Physical damage requiring repair

## Development

```bash
# Local development (mock models)
MOCK_MODELS=true python app.py

# Run tests
pytest tests/ -v
```

## Requirements

- Python 3.10+
- 96GB GPU memory for real model inference (4x L4 or equivalent)
- See `requirements.txt` for full dependencies

## License

Proprietary - For authorized use only.