🍽️ Restaurant Review Intelligence System (RAG-Based)

An end-to-end Retrieval-Augmented Generation (RAG) system that transforms thousands of raw restaurant reviews into trustworthy, evidence-backed insights using LLMs.

Unlike typical AI apps that hallucinate, this system ensures every answer is grounded in real customer reviews.

---

🚀 Why This Project Matters

Choosing a restaurant today means scanning hundreds of reviews across platforms.

This system eliminates that friction by allowing users to ask:

• “Is this place good for business meetings?”
• “Are customers complaining about service delays?”
• “Is the pricing justified by quality?”

…and receive fact-based answers backed by real user reviews.

---

⚡ Key Features

• 🔍 Semantic Search over Reviews (not keyword-based)
• 🧠 RAG Pipeline for grounded responses
• 🎯 Intent-Aware Query Routing (food, pricing, service, ambience)
• 📊 Evidence-Based Answers with source transparency
• 💬 Natural Language Q&A Interface
• ⚡ Real-time interaction via Streamlit

---

🧠 System Architecture

Pipeline Overview:

1. Data Ingestion

   • Yelp Academic Dataset (business + reviews)

2. Preprocessing

   • Restaurant filtering
   • Text cleaning & normalization
   • Metadata linking

3. Embedding Layer

   • Model: mxbai-embed-large
   • Converts reviews → dense vectors

4. Vector Storage

   • ChromaDB for efficient similarity search

5. Retriever

   • Semantic search retrieves top-k relevant reviews

6. Intent Router

   • Classifies query into categories:

     • Food, Service, Price, Ambience, Trends

7. LLM Reasoning

   • Model: LLaMA 3.2 (via Ollama)
   • Generates answers strictly from retrieved context

8. User Interface

   • Streamlit-based interactive dashboard

---

📊 Example Output

🎥 Demo: https://drive.google.com/file/d/1ut47COhxKixWsZ5OpvvXKjOsvhZ8-0qh/view?usp=sharing

---

🧪 What Makes This Different

Most beginner RAG projects:

• Dump documents into a vector DB
• Ask questions → generate answers

This project goes further:

• ✅ Intent-aware routing before generation
• ✅ Structured restaurant filtering (not generic corpus)
• ✅ End-to-end pipeline (data → UI)
• ✅ Focus on real-world decision-making use case

---

⚙️ Tech Stack

Layer	Technology
Language	Python
LLM	LLaMA 3.2 (Ollama)
Embeddings	mxbai-embed-large
Vector DB	ChromaDB
Framework	LangChain
UI	Streamlit
Dataset	Yelp Academic Dataset

---

📌 Limitations

• Performance depends on embedding quality
• No fine-tuning applied (pure RAG pipeline)
• Limited to dataset coverage (no live reviews)

---

🚧 Future Improvements

• Hybrid search (BM25 + embeddings)
• Fine-tuned LLM for better reasoning
• Real-time data ingestion (Google/Yelp APIs)
• Personalized recommendations per user

---

🧠 Key Takeaways

This project demonstrates:

• Practical implementation of RAG systems
• Understanding of LLM limitations (hallucination)
• Ability to design data → retrieval → reasoning pipelines
• Building AI systems with real-world usability