AI
YouTube Summary RAG Video Analyzer
Gemini and FAISS app for transcript-grounded video Q&A
Overview
A RAG video analysis tool that chunks YouTube transcripts, embeds retrieval context with FAISS, and serves interactive summaries and question answering through Streamlit.
Problem
Long videos are difficult to inspect quickly, and generic summaries often miss grounded context from the transcript.
Dataset
YouTube transcript text split into retrievable chunks, with embedded context windows used for user queries.
Approach
Built Gemini Flash LCEL chains, used recursive text splitting, created a FAISS vector store, and designed a Streamlit interface with separate query sections.
Results
Delivered transcript-grounded answers with a four-document retrieval window and a responsive interface for video summaries and targeted questions.
Lessons Learned
RAG quality depends heavily on chunking, retrieval scope, and UI affordances that help users understand where the answer came from.
Model / Pipeline
The implementation combines Python, Google Gemini, LangChain, FAISS, Streamlit with a repeatable workflow for data preparation, evaluation, and communication.