Semantic RAG Chatbot

A Retrieval-Augmented Generation (RAG) chatbot built with Streamlit, FAISS, HuggingFace embeddings, and Google Gemini.
It allows users to upload documents, perform semantic search, and generate accurate, context-aware answers using large language models.

This project also demonstrates RDF-like semantic triple extraction, semantic similarity scoring, and is inspired by research on LLM-powered intelligent knowledge graph construction in medical domains.

---

🚀 Features

• Upload multiple documents
• Extract text from PDF / PPTX / DOCX
• Smart text chunking using RecursiveCharacterTextSplitter
• Vector embeddings with all-MiniLM-L6-v2
• High-speed semantic search using FAISS
• Google Gemini for answer generation
• Auto-generated semantic triples (demo)
• Semantic similarity scoring
• Vector store caching for fast reloads
• Fully interactive Streamlit UI

---

🧠 Tech Stack

• Python 3.10+
• Streamlit
• LangChain
• FAISS
• sentence-transformers
• Google Generative AI (Gemini)

---

📁 Project Structure

muq-bot-rag/
│
├── app.py            # Main Streamlit application
├── utils.py          # Loaders, chunking, FAISS helpers
├── requirements.txt # Dependencies
└── README.md         # Project documentation

---

⚙️ Installation

1. Clone the Repository

git clone https://github.com/mohdmuqsith/muq-bot-rag.git
cd muq-bot-rag

---

2. Create a Virtual Environment

Windows

python -m venv venv
venv\Scripts\activate

Mac / Linux

python3 -m venv venv
source venv/bin/activate

---

3. Install Dependencies

pip install -r requirements.txt

---

🔐 Setting Up API Keys

Option 1: Streamlit Secrets (Recommended)

Create the file:

.streamlit/secrets.toml

Add:

GOOGLE_API_KEY = "your_api_key_here"

---

Option 2: Environment Variable

Windows

setx GOOGLE_API_KEY "your_api_key_here"

Mac / Linux

export GOOGLE_API_KEY="your_api_key_here"

---

▶️ Running the App

streamlit run app.py

Open in your browser:

http://localhost:8501

---

🔍 How It Works

1. User uploads documents
2. Text is extracted
3. Text is chunked
4. Chunks are embedded using HuggingFace
5. Embeddings are stored in FAISS
6. User asks a question
7. Top-k relevant chunks are retrieved
8. Gemini generates the final answer
9. Semantic triples & similarity scores are displayed

---

☁️ Deployment (Streamlit Cloud)

1. Push your repository to GitHub
2. Visit: https://share.streamlit.io
3. Connect your repository
4. Select app.py
5. Add the following in Secrets:

GOOGLE_API_KEY = "your_api_key_here"

6. Deploy 🚀

---

🛠 Troubleshooting

❌ Model not found (404)

Your Gemini account may not support the chosen model.
Use test.py to list available models.

---

❌ FAISS import issues

Install the CPU-only version:

pip install faiss-cpu

---

❌ Streamlit secrets error

Make sure:

• .streamlit/secrets.toml exists
• It is correctly indented
• No extra spaces or quotes

---

🙏 Credits

Inspired by research on LLM-assisted RDF knowledge graph construction, semantic retrieval workflows, and medical ontology mapping using large language models.

Research Inspiration:

• Large Language Models for Intelligent RDF Knowledge Graph Construction
• Published in Frontiers in Artificial Intelligence (2025)
• Focus: Medical ontology mapping, semantic interoperability, and explainable AI

---

👤 Author

Mohammed Abdul Muqsith
Demo Project: https://github.com/mohdmuqsith/muq-bot-rag.git