Local AI Agent

A local AI chatbot that answers questions about a pizza restaurant using customer reviews. Built with LangChain, Ollama, and vector embeddings - runs completely offline with no API costs.

✨ Features

• 100% Local AI - No internet required, no API costs
• RAG (Retrieval-Augmented Generation) - Uses actual review data to answer questions
• Interactive Chat Interface - Ask questions in a conversational loop
• Smart Review Retrieval - Finds the most relevant reviews for each question
• Vector Database - Fast similarity search using Chroma
• Local LLM - Powered by Llama 3.2 1B model

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• Ollama installed and running
• Required Ollama models (see Setup section)

Installation

1. Clone or download this repository

   git clone <your-repo-url>
   cd local-ai-agent

2. Create a virtual environment

   python -m venv venv

3. Activate the virtual environment

   Windows:

   venv\Scripts\activate

   macOS/Linux:

   source venv/bin/activate

4. Install dependencies

   pip install -r requirements.txt

Setup

1. Install and start Ollama

   • Download from ollama.ai
   • Start the Ollama service

2. Pull required models

   ollama pull llama3.2:1b
   ollama pull mxbai-embed-large

3. Ensure you have the data file

   • Make sure realistic_restaurant_reviews.csv is in the project directory
   • The vector database will be created automatically on first run

🎯 Usage

Running the Application

python main.py

Example Interaction

-------------------------------
Ask your question (q to quit): What do people say about the pizza?

The reviews show that people generally enjoy the pizza! Customers mention:
- "The pizza was delicious and fresh"
- "Great variety of toppings"
- "Crust was perfectly crispy"
- "Best pizza in town"

-------------------------------
Ask your question (q to quit): How is the service?

Based on the reviews, the service gets mixed feedback:
- Some customers praise the "friendly and fast service"
- Others mention "slow during busy hours"
- "Staff was helpful and attentive"

-------------------------------
Ask your question (q to quit): q

Sample Questions to Try

• "What do people say about the pizza?"
• "How is the service?"
• "What are the most common complaints?"
• "What do customers like most?"
• "How is the atmosphere?"
• "What about the prices?"

🏗️ Architecture

Components

• main.py - Main application with chat interface
• vector.py - Vector database setup and retrieval
• realistic_restaurant_reviews.csv - Sample restaurant review data
• requirements.txt - Python dependencies

Tech Stack

• LangChain - AI framework for building applications
• Ollama - Local AI model hosting
• Chroma - Vector database for embeddings
• Pandas - Data processing
• Llama 3.2 1B - Local language model
• mxbai-embed-large - Embedding model

How It Works

1. Data Loading - Loads restaurant reviews from CSV
2. Embedding Creation - Converts reviews to vector embeddings
3. Vector Storage - Stores embeddings in Chroma database
4. Query Processing - Converts user questions to embeddings
5. Similarity Search - Finds most relevant reviews (top 5)
6. Context Generation - Passes relevant reviews to LLM
7. Answer Generation - LLM generates answer based on context

📁 Project Structure

local-ai-agent/
├── main.py                          # Main application
├── vector.py                        # Vector database setup
├── config.py                        # Configuration settings
├── requirements.txt                 # Python dependencies
├── realistic_restaurant_reviews.csv # Sample data
├── chrome_langchain_db/             # Vector database (auto-created)
└── README.md                        # This file

🔧 Configuration

Model Settings

• LLM Model: llama3.2:1b (lightweight, fast)
• Embedding Model: mxbai-embed-large (high-quality embeddings)
• Retrieval Count: 5 most relevant reviews per question

Customization

You can modify all settings in config.py:

# Model Configuration
LLM_MODEL = "llama3.2:1b"                   # Change LLM model
EMBEDDING_MODEL = "mxbai-embed-large"       # Change embedding model

# Vector Database Configuration
RETRIEVAL_COUNT = 5                          # Number of reviews to retrieve

# Data Configuration
CSV_FILE = "realistic_restaurant_reviews.csv" # Change data source

Available Models:

• LLM Models: llama3.2:1b, llama3.2:3b, llama3.1:8b, llama3.1:70b
• Embedding Models: mxbai-embed-large, nomic-embed-text, all-minilm

🐛 Troubleshooting

Common Issues

1. "Model not found" error

   • Ensure Ollama is running: ollama serve
   • Pull required models: ollama pull llama3.2:1b

2. "Chroma database" errors

   • Delete the chrome_langchain_db folder and restart
   • The database will be recreated automatically

3. Slow performance

   • The first run is slower (building vector database)
   • Subsequent runs are much faster

4. Memory issues

   • Try a smaller model: ollama pull llama3.2:1b
   • Reduce retrieval count in vector.py

Performance Tips

• First run: Takes longer to build vector database
• Subsequent runs: Much faster (database is cached)
• Model size: Llama 3.2 1B is optimized for speed
• Retrieval: Limited to 5 reviews for fast responses

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Test thoroughly
5. Submit a pull request

📄 License

This project is open source and available under the MIT License.

🔗 Links

• Ollama: https://ollama.ai/
• LangChain: https://python.langchain.com/
• Chroma: https://www.trychroma.com/

---

Built with ❤️ for local AI experimentation