🎥 Pragya - YouTube AI Assistant

A full-stack AI-powered application that summarizes and answers questions in many different languages from YouTube videos using Azure OpenAI, Whisper, and LangChain.

🚀 Getting Started

🔧 Backend Setup

Set environment variables
Create a .env file or export manually with the following keys and get the Api keys from azure Openai:

AZURE_OPENAI_API_KEY=
AZURE_OPENAI_ENDPOINT=
AZURE_DEPLOYMENT_NAME=
AZURE_WHISPER_API_KEY=
AZURE_WHISPER_ENDPOINT=
AZURE_EMBEDDING_DEPLOYMENT_NAME=
AZURE_EMBEDDING_DEPLOYMENT_VERSION=
AZURE_INFERENCE_ENDPOINT=
AZURE_EMBEDDING_API_KEY=
AZURE_WHISPER_VERSION=
AZURE_TTS_API_KEY=
AZURE_TTS_ENDPOINT=

Run the backend locally

python -m venv venv
Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy Unrestricted  # (PowerShell only)
.\venv\Scripts\Activate.ps1                                           # (PowerShell)
pip install -r requirements.txt
python main.py

🌐 Frontend Setup

npm install
npm run dev

Demo

🛠️ Tech Stack

✅ Backend

Flask + LangChain + Azure OpenAI + Whisper
- Accepts YouTube links and user queries
- Retrieves transcripts via YouTube Transcript API or Whisper
- Uses GPT-4o-mini for summarization and question answering

✅ Vector Store

FAISS + LangChain
- Splits transcripts into chunks and embeds them
- Stores embeddings for semantic search
- Retrieves relevant chunks for QA

✅ Chat Memory

LangChain Memory Modules
- Maintains dialogue history
- Supports contextual follow-up questions and coreference resolution

✅ Frontend

React
- UI for submitting YouTube links
- Choose between Summary or QnA modes
- Select language for response
- Optionally listen to responses via Azure TTS

📦 Features

Summarize or query any YouTube video
Auto transcript via Whisper fallback
Semantic retrieval with FAISS
Memory-enabled multi-turn conversations
Audio output with Azure TTS

FilesExpand file tree

README.md

Latest commit

History