🧠 RAG-Based Teaching AI Assistant

🎓 A complete Retrieval-Augmented Generation (RAG) pipeline that transforms video lectures into an AI-powered teaching assistant, using offline speech recognition (Vosk) and context-aware Q&A through Groq or Gemini.

---

🚀 Overview

This project turns your teaching videos into an intelligent assistant capable of answering questions based on their content.
It combines offline transcription, semantic embeddings, and LLM reasoning to make learning interactive and accessible anywhere — even offline.

---

🧩 Key Features

✅ Video → Audio → Text Pipeline
Convert .mp4 lectures into .mp3 and transcribe them using Vosk (offline ASR).

✅ Clean JSON Output
Automatic cleaning of transcripts for easy processing and embedding.

✅ Embeddings + FAISS Index
Generate semantic embeddings with SentenceTransformers and store them in a FAISS vector database for lightning-fast retrieval.

✅ RAG with Groq or Gemini
Retrieve relevant transcript chunks and answer questions with context-aware reasoning using Groq or Google Gemini APIs.

✅ Interactive Console
Ask natural-language questions directly inside the notebook.

---

🧱 Project Structure

📦 RAG-Based-AI-Teaching-Assistant/
├── videos/               # Original lecture videos (.mp4)
├── audios/               # Converted audio files (.mp3)
├── jsons/                # Raw Vosk transcripts
├── jsons/clean/          # Cleaned transcripts
├── vosk_model/           # Offline Vosk model directory
├── embeddings.joblib     # Stored embeddings for retrieval
├── RAG_Teaching_AI_Assistant.ipynb  # Main Colab notebook
└── README.md             # Project documentation

---

⚙️ Setup & Installation

1️⃣ Clone or open in Google Colab

git clone https://github.com/<your-username>/RAG-AI-Teaching-Assistant.git
cd RAG-AI-Teaching-Assistant

or open directly in Colab and run all cells sequentially.

2️⃣ Install dependencies

pip install vosk soundfile pydub tqdm ffmpeg-python sentence-transformers joblib pandas faiss-cpu groq google-generativeai

Ensure that ffmpeg is installed (Colab auto-installs it).

---

🔄 Workflow Summary

Step	Description	Notebook Cell
1	Install & configure Vosk	Cell 1
2	Mount Google Drive & import project	Cell 2
3	Ensure required folders exist	Cell 3
4	Convert videos → MP3	Cell 4
5	Transcribe audio → JSON (offline)	Cell 5
6	Clean & preprocess transcripts	Cell 6
7	Generate & store embeddings	Cell 7
8	Build FAISS index + RAG QA (Groq/Gemini)	Cell 8

---

🔑 API Configuration

Set up one of the following API keys before running Cell 8:

🧠 Using Groq

import os
os.environ["GROQ_API_KEY"] = "your_groq_api_key"

🌐 Using Gemini

import os
os.environ["GEMINI_API_KEY"] = "your_gemini_api_key"

---

💬 Example Interaction

🔌 Provider: GROQ
❓ Your question: What is a black hole?

🔎 Retrieved Context:
 [1] Black holes are regions in space where gravity is so strong...

💬 Answer:
 A black hole is a region in space where the gravitational pull is so intense that nothing, not even light, can escape.

---

🧠 Tech Stack

Category	Tools / Libraries
Language	Python 3.10+
Environment	Google Colab
Speech-to-Text	Vosk
Audio Processing	FFmpeg, Pydub
Embeddings	Sentence-Transformers
Vector Search	FAISS
LLM API	Groq / Google Gemini
Data Serialization	JSON, Joblib, Pandas

---

📈 Future Enhancements

• 🔊 Integrate Whisper or OpenAI ASR as optional transcription backend
• 🌍 Add multilingual transcription and embedding support
• 🧩 Build a Gradio/Streamlit UI for web-based chat
• 🧮 Summarize or cluster lectures by topic
• 💾 Integrate vector-store persistence (e.g., ChromaDB)

---

👨‍💻 Author

Mohammed Enayatullah Safwan 🎓 National Institute of Technology, Durgapur 📧 LinkedIn | GitHub

✨ “Transform your lectures into knowledge-driven conversations — powered by AI.”