Multilingual YouTube RAG Assistant

Problem

Educational and research users struggle to extract insights from long multilingual videos because keyword search misses context and watching full videos is time-consuming.

Solution

Implemented a Flask-based RAG application that embeds YouTube transcripts with FAISS and Hugging Face multilingual models, then uses Gemini 2.5 Flash to answer questions, summarize content, and maintain multi-turn session context.

Highlights

• Semantic retrieval over transcript segments instead of keyword-only search
• Multilingual Q&A and summarization in one interface
• Session-aware multi-turn conversations for deeper exploration

Outcomes

• Faster comprehension of long-form video content
• More accurate responses through retrieval-grounded generation
• Reusable architecture for education and content-analysis workflows