README.md

---
emoji: 👀
colorTo: purple
colorFrom: purple
---
# 🧠 YouTube Transcript Chatbot (RAG) — LLaMA, Gemma, DeepSeek, OpenAI

An advanced **Retrieval-Augmented Generation (RAG)** chatbot that answers questions from YouTube video transcripts using top open-source LLMs — **Llama 3.2**, **Gemma 2**, **DeepSeek**, and **OpenAI GPT-OSS**.  

Built with **Streamlit**, **LangChain**, **Hugging Face Hub**, and **FAISS**.

---

## 🚀 Features

✅ Fetch YouTube transcripts via **RapidAPI**  
✅ Embed and store transcript chunks using **FAISS**  
✅ Query understanding through **LangChain retrieval pipeline**  
✅ Supports **multiple Hugging Face chat-based models**:
- 🦙 **Llama-3.2-1B-Instruct**
- 💎 **Gemma-2-2B-it**
- 🧠 **DeepSeek-V3.2-Exp**
- 🔮 **OpenAI GPT-OSS-20B**

✅ Adjustable **temperature** for creative or factual outputs  
✅ Seamless **Streamlit UI**  
✅ Caching for faster repeat queries  

---

## 🧩 Architecture


---

## 🧠 Tech Stack

| Component | Technology |
|------------|-------------|
| Frontend | Streamlit |
| Backend | LangChain |
| LLMs | Hugging Face Endpoints |
| Embeddings | intfloat/multilingual-e5-base |
| Vector Store | FAISS |
| API | YouTube Transcript API (RapidAPI) |

---