AI YouTube Video Summarizer (Laravel)
A real-time content summarization tool for YouTube transcripts with WebSocket streaming, content chunking, and support for multiple AI models.
- Laravel
- PHP
- WebSockets
- AI APIs
- Real-time streaming
Problem
Long-form video hides useful information behind hours of watch time. Generic summaries often lack depth and give no path to ask follow-up questions.
Solution
A Laravel application that fetches transcripts, chunks content for reliable model context, streams summaries over WebSockets, and supports multi-model follow-ups.
My role
Full-stack engineering - real-time streaming, AI integration, transcript processing, and the summary experience.
Impact
Turns long videos into structured insight quickly, with streaming feedback and room to explore further through questions.
Challenges
Keeping streaming reliable under variable model latency, chunking long transcripts without losing coherence, and designing a follow-up flow that stays useful.
Screenshots
The problem
Video is rich but slow to consume. For research, learning, and team knowledge sharing, watching every minute is rarely the best use of time - yet shallow summaries often miss the nuance people actually need.
The approach
Content Summariser treats summarization as an interactive product, not a one-shot API call:
- Fetch and prepare YouTube transcripts
- Chunk long content so models stay within reliable context windows
- Stream summary output in real time via WebSockets
- Support follow-up questions and multiple AI models
Architecture notes
Laravel owns orchestration: transcript handling, chunking strategy, model calls, and streaming to the client. Real-time delivery keeps the experience responsive even when generation takes time.
Trade-offs
Streaming improves perceived performance but requires careful handling of partial output and failures. Multi-model support adds flexibility at the cost of normalizing different provider behaviors. Chunking improves reliability while demanding thoughtful stitching of context across sections.
Outcome
The result is a practical tool for extracting signal from long-form video - with enough depth to continue the conversation after the first summary.