Repair Advisor RAG System

A Retrieval-Augmented Generation (RAG) system for providing intelligent repair advice and assessments based on a large database of repair cases. The system uses OpenAI's GPT models for generation and Pinecone for efficient vector similarity search.

Overview

The system combines:

• OpenAI GPT Models for generating repair advice
• Pinecone Vector Database for efficient similarity search
• OpenRepair Dataset as the knowledge source
• Modern Web Interface for easy interaction

Quick Start

1. Clone & Install

   git clone <repository-url>
   cd repair_rag
   pip install -r requirements.txt

2. Configure Environment Create a .env file in the repair_rag directory:

   # OpenAI Configuration
   OPENAI_API_KEY=your_openai_api_key_here
   OPENAI_MODEL=gpt-4o-mini-2024-07-18
   EMBEDDING_MODEL=text-embedding-3-small
   
   # Pinecone Configuration
   PINECONE_API_KEY=your_pinecone_api_key_here
   PINECONE_ENVIRONMENT=gcp-starter
   PINECONE_INDEX=repair-kb
   PINECONE_NAMESPACE=repair-cases
   

3. Start the API Server

   cd repair_rag
   python run.py api

4. Access the Web Interface

   • Open index.html in your browser
   • Start asking repair-related questions!

Project Structure

repair_rag/
├── app/                    # Core application code
│   ├── main.py            # FastAPI server
│   ├── rag_engine.py      # RAG implementation
│   └── __init__.py        # Package marker
├── data/                   # Data storage
├── index.html             # Main web interface
├── results.html           # Results display
├── styles.css             # UI styling
├── run.py                 # Application launcher
├── ingest_data_to_pinecone.py  # Data ingestion
├── parts_url_generator.py      # Parts URL utilities
└── requirements.txt       # Python dependencies

Features

• Intelligent Repair Analysis: Combines vector similarity search with GPT-powered analysis
• Comprehensive Advice:
  • Repair feasibility assessment
  • Cost-benefit analysis
  • Required tools and parts
  • Step-by-step guidance
• Multi-language Support: Automatically detects and responds in the user's language
• Similar Case Analysis: Shows relevant past repair cases
• Parts & Tools Links: Direct links to purchase necessary items

Adding Your Own Data

To add new repair cases to the existing database:

1. Prepare Your Data Create a JSON file with the following structure:

   [
     {
       "brand": "Device brand",
       "product_category": "Category of the device",
       "problem": "Description of the problem",
       "repair_status": "Fixed/Not Fixed/etc.",
       "repair_barrier": "Any barriers to repair",
       "year_of_manufacture": "Year",
       "country": "Country",
       "tools_required": ["Tool1", "Tool2"],
       "parts_needed": ["Part1", "Part2"],
       "estimated_cost": "Cost estimate",
       "repair_description": "Detailed repair description"
     },
     // ... more cases
   ]

2. Ingest to Pinecone

   python ingest_data_to_pinecone.py --data-path your_data.json

   Options:

   • --batch-size: Number of vectors to batch (default: 50)
   • --embedding-model: OpenAI embedding model to use
   • --index-name: Pinecone index name
   • --namespace: Namespace within the index

Troubleshooting

Common Issues

1. API Key Errors

   • Ensure both OpenAI and Pinecone API keys are in .env
   • Check API key permissions and quotas

2. Connection Issues

   • Verify the API server is running (python run.py api)
   • Check the port (8000) isn't in use

3. No Results

   • Ensure Pinecone index contains data
   • Check query formatting and language

Rate Limits & Optimization

• OpenAI has rate limits for both embeddings and completions
• Pinecone has query per second (QPS) limits
• Use batch processing for large data ingestion
• Consider implementing caching for frequent queries

Technical Details

Vector Database

• Uses Pinecone for vector similarity search
• Dimension: 1536 (OpenAI text-embedding-3-small)
• Metric: Cosine similarity
• Namespace: Organizes different data sources

Language Models

• Embeddings: text-embedding-3-small
• Generation: gpt-4o-mini-2024-07-18
• Temperature: 0.1 (for factual responses)

Web Interface

• Pure HTML/CSS/JavaScript
• No framework dependencies
• Responsive design
• Cross-browser compatible

Pricing Considerations

1. OpenAI Costs:

   • Embeddings: $0.02 per million tokens
   • GPT-4 Completions: Varies by model and usage

2. Pinecone Costs:

   • Free tier available
   • Pricing based on vector count and QPS

License & Attribution

• Built on OpenRepair Dataset
• Uses OpenAI's API
• Pinecone Vector Database
• [Add your license information]