AutoTechQA

AutoTechQA is an advanced AI-powered question-answering system designed to assist users in finding accurate and relevant information from automotive truck and bus technical manuals. Built using RAG (Retrieval-Augmented Generation) architecture, AutoTechQA leverages state-of-the-art natural language processing and vector search to understand complex queries and provide precise answers from technical documentation.

Features

• 🚛 Truck & Bus Manual QA: Specialized in automotive technical documentation
• 🔍 Semantic Search: Uses vector embeddings for intelligent document retrieval
• 🤖 GPT-4 Powered: Leverages OpenAI's GPT-4 for accurate answer generation
• 💾 Vector Database: ChromaDB for efficient similarity search
• 🎨 User-Friendly Interface: Streamlit web application for easy interaction
• 📄 PDF Ingestion: Automated processing of PDF technical manuals

Architecture

The system follows a RAG (Retrieval-Augmented Generation) pipeline:

1. Document Ingestion (ingest.py): PDF manuals are split into chunks and embedded using HuggingFace embeddings
2. Vector Storage (ChromaDB): Document chunks stored in vector database for efficient retrieval
3. Query Processing (rag_core.py): User questions are used to retrieve relevant context
4. Answer Generation: GPT-4 generates answers based on retrieved context
5. Web Interface (app.py): Streamlit UI for user interaction

Project Structure

AutoTechQA/
├── src/
│   ├── app.py          # Streamlit web application
│   ├── config.py       # Configuration parameters
│   ├── ingest.py       # PDF ingestion and vector store creation
│   └── rag_core.py     # RAG chain implementation
├── data_source/        # Place PDF manuals here
├── vector_store/       # ChromaDB vector databases
├── requirements.txt    # Python dependencies
└── README.md

Configuration

The system can be customized via src/config.py:

• LLM Model: gpt-4 (default)
• Embedding Model: nomic-ai/nomic-embed-text-v1
• Chunk Size: 1000 characters
• Chunk Overlap: 100 characters
• Top-K Retrieval: 5 documents

Installation

1. Clone the repository

   git clone https://github.com/indhra/AutoTechQA.git
   cd AutoTechQA

2. Create virtual environment

   python -m venv .venv
   source .venv/bin/activate  # On macOS/Linux
   # or
   .venv\Scripts\activate     # On Windows

3. Install dependencies

   pip install -r requirements.txt

4. Set up environment variables

   Create a .env file in the project root:

   OPENAI_API_KEY=your_openai_api_key_here
   

Usage

1. Ingest PDF Manuals

Place your PDF manuals in the data_source/ folder, then run:

python src/ingest.py

This will:

• Load PDFs from data_source/
• Process the first 30 pages of each manual
• Split text into chunks (1000 characters with 100 overlap)
• Create vector embeddings using HuggingFace nomic-embed model
• Store vectors in ChromaDB in vector_store/
• Skip already processed files by default

2. Run the Application

Start the Streamlit web interface:

streamlit run src/app.py

The application will be available at http://localhost:8501

3. Ask Questions

Enter your questions about trucks or buses in the web interface. The system will:

• Retrieve the 5 most relevant document chunks
• Use GPT-4 to generate an accurate answer
• Display the response in the UI

Technical Stack

• LangChain: Framework for LLM applications and RAG pipelines
• OpenAI GPT-4: Large language model for answer generation
• ChromaDB: Vector database for similarity search
• HuggingFace Transformers: Embedding models (nomic-embed-text-v1)
• Streamlit: Web application framework
• PyPDF: PDF parsing and text extraction

Dependencies

Key dependencies (see requirements.txt for complete list):

• streamlit
• langchain_core, langchain_openai, langchain_community
• langchain_huggingface, langchain-chroma
• chromadb
• sentence-transformers
• pypdf
• python-dotenv

Notes

• The system currently processes only the first 30 pages of each PDF manual
• Existing vector stores are skipped during re-ingestion to save processing time
• All answers are strictly based on the provided context from manuals
• If information is not found in the manuals, the system explicitly states this

Future Enhancements

Potential improvements:

• Support for more document formats
• Multi-language support
• Advanced filtering and metadata search
• Conversation history and context
• Source citation with page numbers
• Processing full PDFs instead of first 30 pages

License

MIT License

Contact

N A INDHRA KIRANU