🧠 Phi RAG Chatbot

A lightweight Retrieval-Augmented Generation (RAG) chatbot that uses Ollama's Phi model with a custom document index to provide context-aware answers. Built with modularity, scalability, and simplicity in mind.

---

📂 Project Structure

rag-phi-bot/
├── chatbot/
│   ├── chatbot.py          # Main chatbot loop and chat logic
│   ├── phi_interface.py    # Stream interface for Ollama's Phi model
│   ├── document_indexer.py # PDF parser, chunker, embedder, FAISS indexer
│   ├── retriever.py        # Loads index, retrieves top-matching chunks
│   └── utils.py            # Hallucination filter, keyword loader
├── data/
│   └── AkhilSuresh_CV.pdf  # Example PDF to index
├── doc_index/              # Auto-generated FAISS index & metadata
│   └── faiss.index
│   └── metadata.pkl
├── hallucination_keywords.txt # Optional custom hallucination keywords
├── run_chat.py             # Chat CLI entry point
├── index_document.py       # Script to index a new PDF
├── requirements.txt
└── README.md

---

🚀 Features

• 🔎 RAG-based responses: Embeds and indexes PDF text for retrieval.
• 💬 Streaming chat: Real-time streaming replies using Ollama’s phi model.
• 📄 PDF Support: Upload your own documents for indexing.
• 🧠 Hallucination Guard: Auto-aborts hallucinated responses using keyword detection.
• 🧱 Modular Design: Easily extensible and maintainable codebase.

---

🛠️ Installation

1. Install Ollama and run the Phi model:

   curl -fsSL https://ollama.com/install.sh | sh
   ollama run phi

2. Clone and set up this repo:

   git clone "https://github.com/akhilpsin/rag-phi-bot.git"
   cd rag-phi-bot
   pip install -r requirements.txt

3. Index your documents:

   Place your PDF in the data/ folder, then run :

   python index_document.py

4. Start chatting:

   python run_chat.py

---

📦 Dependencies

Install via:

pip install -r requirements.txt

Key Libraries:

• faiss-cpu
• sentence-transformers
• PyMuPDF (fitz)
• ollama
• aiofiles
• numpy, pickle

---

Output Screenshots

1. With Document Mode ON

When Document Mode is enabled, the chatbot retrieves and uses relevant documents to provide more accurate answers. Here is how the chatbot behaves with this feature enabled:

2. With Document Mode OFF

When Document Mode is disabled, the chatbot relies only on its internal knowledge base to answer the questions. Here is how the chatbot behaves without this feature:

---

Instructions:

• Ensure that you have enabled the proper mode when starting the chatbot (Y for Document Mode ON, N for Document Mode OFF).
• Check the output screenshots to see the difference in how the bot answers with or without using documents.

📌 Notes

• To customize hallucination filtering, edit hallucination_keywords.txt.
• Embeddings are generated using all-MiniLM-L6-v2 from SentenceTransformers.
• Only English-language PDFs are supported for now.