RAG-Based AI Agent

A Retrieval-Augmented Generation (RAG) chatbot that answers natural language questions about a pizza restaurant using real customer reviews, powered by Ollama.

---

How It Works

User Question
     │
     ▼
ChromaDB Vector Store  ──►  Top-K relevant reviews
     │
     ▼
Ollama LLM (phi3)  ──►  Answer grounded in real reviews

1. Ingestion - Customer reviews from data/restaurant_reviews.csv are embedded using mxbai-embed-large and stored in a local ChromaDB vector store (built once, reused every run).
2. Retrieval - For each question, the 5 most semantically relevant reviews are fetched.
3. Generation - A local LLM (phi3) reads those reviews and answers the question.

---

Prerequisites

• Python 3.9+
• Ollama installed and running

Pull the required models:

ollama pull phi3
ollama pull mxbai-embed-large

---

Setup

# 1. Clone the repo
git clone https://github.com/YOUR_USERNAME/restaurant-review-agent.git
cd restaurant-review-agent

# 2. Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Run
python main.py

On the first run, the vector store is built automatically from the CSV. Subsequent runs reuse the cached store and start instantly.

---

Example

   Restaurant Review AI Agent
   Powered by Ollama + phi3
   Type 'q' or 'quit' to exit

--------------------------------------------------
Ask a question about the restaurant: Is the pizza worth the price?

Answer: Based on the reviews, opinions are mixed. Several customers found the
pizza exceptional and worth every penny, praising the quality of ingredients
and wood-fired crust. However, a few reviewers felt the pricing was high
for the portion size. Overall, the majority lean positive on value.

---

Project Structure

restaurant-review-agent/
├── scripts/
│   ├── main.py                       # CLI entry point & conversation loop
│   └── vector.py                     # Vector store setup and retriever
│   └── config.py                     # Centralized configuration
├── data/
│   └── restaurant_reviews.csv  # Source reviews dataset
├── requirements.txt
├── .gitignore
└── README.md

---

Configuration

All settings live in config.py:

Setting	Default	Description
LLM_MODEL	phi3	Ollama model for answer generation
EMBEDDING_MODEL	mxbai-embed-large	Ollama model for embeddings
CSV_PATH	data/restaurant_reviews.csv	Path to reviews dataset
DB_LOCATION	./chroma_langchain_db	ChromaDB persistence directory
RETRIEVER_K	5	Number of reviews retrieved per query

You can swap in any Ollama-compatible model - for example, replace phi3 with llama3 for more capable responses.

---

Tech Stack

Component	Library / Tool
LLM	Ollama (phi3)
Embeddings	Ollama (mxbai-embed-large)
Vector Store	ChromaDB
Orchestration	LangChain
Data	Pandas

---

License

MIT