🤖 RAG AI Chatbot — Powered by Gemini & MongoDB Atlas

A full-stack Retrieval-Augmented Generation (RAG) system that lets you upload your own knowledge base and chat with it using Google Gemini. Built with LangChain, MongoDB Atlas Vector Search, and Streamlit.

---

Screenshots

💬 Chatbot Homepage

🗄️ MongoDB Vector Data

📊 Semantic Distance Map (Vector Visualization)

---

Features

• Vector Store — MongoDB Atlas Vector Search stores and retrieves document embeddings
• Embeddings — Powered by sentence-transformers/all-mpnet-base-v2 (768 dimensions)
• LLM — Google Gemini gemini-2.5-flash for natural language responses
• Vector Visualization — PCA-based 2D semantic distance map using Plotly
• Framework — Built with LangChain + Streamlit

---

Project Structure

rag_template/
├── .streamlit/
│   └── secrets.toml        # API keys (never commit this!)
├── assets/
│   ├── chatbot_homepage.png
│   ├── mongo_db_data.png
│   └── semantic_distance_map.png
├── pages/
│   └── vector_graph.py     # Vector visualization page
├── backend.py              # Core RAG logic
├── home.py                 # Streamlit chat UI
└── requirements.txt

---

Prerequisites

1. Python 3.8+
2. A MongoDB Atlas cluster with Vector Search enabled
3. A Google AI Studio API key (Gemini)

---

Installation

1. Clone the repository:

git clone <your-repo-url>
cd rag_template

2. Create and activate a virtual environment:

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

3. Install dependencies:

pip install -r requirements.txt

---

Configuration

Create a .streamlit/secrets.toml file in the project root:

MONGO_URI = "mongodb+srv://<username>:<password>@<cluster>.mongodb.net/?retryWrites=true&w=majority"
GEMINI_API_KEY = "your-gemini-api-key"

⚠️ Never commit secrets.toml to GitHub. Make sure .gitignore includes .streamlit/secrets.toml

---

MongoDB Atlas Setup

1. Create a collection: vector_store_database.embeddings_stream
2. Create a Vector Search Index named vector_index with this definition:

{
  "fields": [
    {
      "numDimensions": 768,
      "path": "embedding",
      "similarity": "cosine",
      "type": "vector"
    }
  ]
}

---

How It Works

User Input Text
      ↓
HuggingFace Embeddings (768-dim vectors)
      ↓
MongoDB Atlas Vector Store
      ↓
User Query → Similarity Search (top 3 docs)
      ↓
Context + Query → Gemini LLM
      ↓
Answer + Sources

---

Key Functions (backend.py)

Function	Description
get_vector_store()	Connects to MongoDB and loads the embedding model
ingest_text(text)	Converts text to vector and stores in MongoDB
get_rag_response(query)	Retrieves top 3 similar docs and generates a Gemini answer
get_vectors_for_visualization(query)	Returns vectors for PCA plotting

---

Run the App

streamlit run home.py

Open http://localhost:8501 in your browser.

1. Paste your knowledge in the sidebar → click Upload to MongoDB
2. Ask questions in the chat input
3. Visit the Vector Visualization page to explore semantic similarity

---

Tech Stack

Tool	Purpose
Streamlit	Frontend UI
LangChain	RAG pipeline orchestration
MongoDB Atlas	Vector store
Google Gemini	Language model
HuggingFace	Sentence embeddings
Plotly + PCA	Vector visualization

---

📄 License

MIT License