Youtube_RAG

An AI-powered YouTube Content Synthesizer using RAG (Retrieval-Augmented Generation). Built with Streamlit, LangChain, and Gemini 2.0 to transform video transcripts into structured notes, important topics, and an interactive chatbot.

VidSynth AI: YouTube Content Synthesizer

VidSynth AI is a sophisticated RAG-based application that allows users to "talk" to YouTube videos and extract high-quality summaries. By leveraging Gemini 2.0 Flash and Vector Databases (ChromaDB), it provides an intelligent way to consume video content efficiently.

Key Features

1. Automated Note-Taking: Generates well-structured, bulleted notes from any video transcript.

2. Topic Extraction: Identifies and summarizes the 5 most important technical points from the video.

3. Interactive Chatbot: A RAG-powered interface to ask specific questions about the video content.

4. Multi-lingual Support: Automatically translates non-English transcripts for analysis.

Tech Stack

1. Framework: Streamlit

2. Orchestration: LangChain

3. AI Models: Gemini 2.0 Flash (Chat) & Text-Embedding-004 (Embeddings)

4. Vector Store: ChromaDB

Technical Challenges & Disclaimers

This project was developed as a technical challenge to implement RAG (Retrieval-Augmented Generation) on a cloud environment. Due to the nature of free-tier cloud hosting and external APIs, users should be aware of the following:

1. YouTube Request Blocking (Proxy Integration)

YouTube frequently blocks automated requests (transcription fetching) from cloud providers like Streamlit.

The Solution: I have integrated a Webshare Proxy Server to route requests through a rotating IP.

The Limitation: Since this uses a free-tier proxy, YouTube may still occasionally block the connection. If you see a "Fetching Transcript Error," it is likely due to the proxy being temporarily throttled by YouTube's security layers.

2. Google Gemini API Quota (429 Errors)

The "Chat with Video" feature requires significant API credits to create embeddings for the vector database.

The Limitation: Users may encounter a 429 RESOURCE_EXHAUSTED error. This is a provider-side quota limit of the Google Gemini Free Tier and not a bug in the source code.

Recommendation: The "Notes For You" feature uses fewer credits and is more likely to work consistently. For chatting, please use shorter videos (under 5 minutes).

Performance Optimizations

To make this work on Streamlit Cloud, I implemented the following:

1. pysqlite3-binary: Used to override the system SQLite version, as Streamlit's default environment is often incompatible with the latest ChromaDB requirements.

2. Proxy Authentication: Configured youtube_transcript_api to use authenticated proxy headers to bypass IP-based blocks.