All Projects
RAGCited Q&AEducation & EdTechMedia

YouTube Chatbot — RAG with Timestamped Citations

A RAG chatbot that lets users chat with YouTube videos and playlists, returning cited, timestamped answers.

Build a Similar AI SystemWatch Demo
YouTube Chatbot — RAG with Timestamped Citations — RAG preview
RAG

Workflow & Architecture

How this system works

Step 1
Video ingestion

YouTube URLs and playlists ingested; captions extracted and cleaned.

Step 2
Chunk & index

Transcripts chunked and stored in a FAISS vector index for retrieval.

Step 3
Query & retrieve

User question matched to the most relevant transcript segments.

Step 4
Cited answer

LLM synthesizes a response with clickable timestamp citations.

Overview

Captions are cleaned, chunked, and indexed for retrieval. Single-video queries return timestamped mini-answers; multi-video queries are merged into one synthesized reply, each backed by clickable citations that jump to the exact moment in the source video.

The Problem

Long-form video content is hard to search — users need specific answers, not full re-watches.

The Solution

Built a RAG system with FAISS indexing, smart chunking, multi-video synthesis, and clickable timestamp citations — plus mid-conversation ability to add new video links.

Key Features

  • Citation-backed answers with clickable timestamp jumps
  • Multi-video synthesis vs. single-video timestamped answers
  • Mid-conversation ability to add new video links and keep chatting
  • "Used" highlights showing which transcript sections informed the answer
  • React + TypeScript frontend

Outcomes & Business Value

  • Citation-grounded RAG pattern for any long-form content library
  • Faster video knowledge search with timestamp navigation
  • Applicable to courses, training libraries, and podcasts

Related Projects