Engineering Drawing Analyzer

An AI-powered tool to analyze engineering drawings using open-source vision models (Ollama + LLaVA). Load drawings, ask questions, and get detailed analysis - all running locally on your machine!

Features

• 🎯 Analyze engineering drawings with natural language prompts
• 🖼️ Support for single or multiple drawings
• 🔒 100% Local & Private - runs entirely on your machine using open-source models
• 💰 Free - no API costs or subscriptions
• 🚀 Multiple vision models supported (LLaVA, BakLLaVA, Moondream)
• 📊 Extract dimensions, specifications, and technical details
• 🔍 Identify components, materials, and design features
• ⚡ Interactive mode for multiple queries

Prerequisites

1. Install Ollama

Ollama is required to run the open-source vision models locally.

macOS:

brew install ollama

Linux:

curl -fsSL https://ollama.ai/install.sh | sh

Windows: Download from https://ollama.ai

2. Start Ollama Service

ollama serve

3. Pull a Vision Model

# Recommended: LLaVA (best for engineering drawings)
ollama pull llava

# Alternative models:
# ollama pull bakllava    # Good for detailed analysis
# ollama pull moondream   # Lightweight and fast

Installation

1. Clone or download this repository
2. Install Python dependencies:

pip install -r requirements.txt

Usage

Command Line Mode

Analyze a single drawing:

python main.py path/to/drawing.png -p "What are the main dimensions shown in this drawing?"

Analyze multiple drawings:

python main.py drawing1.png drawing2.png -p "Compare these two designs"

Use a different model:

python main.py drawing.png -p "List all components" -m bakllava

Interactive Mode

python -c "from main import interactive_mode; interactive_mode()"

Then enter your drawing path(s) and ask multiple questions interactively.

Python API

from drawing_analyzer import DrawingAnalyzer

# Initialize analyzer
analyzer = DrawingAnalyzer(model="llava")

# Analyze a drawing
result = analyzer.analyze_drawing(
    "path/to/drawing.png",
    "What materials are specified in this drawing?"
)
print(result)

# Analyze multiple drawings
result = analyzer.analyze_multiple_drawings(
    ["drawing1.png", "drawing2.png"],
    "What are the differences between these designs?"
)
print(result)

Example Prompts

• "What are the main dimensions in this engineering drawing?"
• "List all the components shown in this assembly"
• "What material specifications are indicated?"
• "Identify any tolerances or technical requirements"
• "What type of engineering drawing is this (orthographic, isometric, etc.)?"
• "Extract the title block information"
• "Are there any welding symbols or surface finish specifications?"

Supported Image Formats

• PNG
• JPEG/JPG
• BMP
• GIF
• TIFF

Available Models

Model	Size	Speed	Quality	Best For
llava	~4.5GB	Medium	High	General engineering drawings
bakllava	~4.5GB	Medium	Very High	Detailed technical analysis
moondream	~1.6GB	Fast	Good	Quick reviews, lighter workloads

Configuration

You can set environment variables in a .env file:

OLLAMA_MODEL=llava
OLLAMA_BASE_URL=http://localhost:11434

Troubleshooting

"Cannot connect to Ollama"

• Ensure Ollama is running: ollama serve
• Check if the service is accessible: curl http://localhost:11434/api/tags

Model not found

• Pull the model first: ollama pull llava
• Check available models: ollama list

Slow performance

• Try a smaller model like moondream
• Ensure you have adequate RAM (8GB+ recommended)
• Close other resource-intensive applications

Requirements

• Python 3.8+
• Ollama installed and running
• At least one vision model pulled (llava, bakllava, or moondream)
• 8GB+ RAM recommended

License

MIT License - feel free to use and modify!

Privacy & Security

All analysis happens locally on your machine. No data is sent to external servers, ensuring your engineering drawings remain private and secure.