VidWhiz

Chat with YouTube Videos

StatusLive & Growing
Timeline3 months
RoleProduct · Full-Stack · DevOps
TypePersonal Product
Tech Stack
Next.jsOpenAIRAGAzureChrome Extension

Built a Chrome extension + backend service that lets users interact with YouTube videos using natural language, powered by RAG and OpenAI APIs.

The Problem

I found myself wasting 30–40 minutes watching YouTube videos just to find one recommendation or answer. I thought: what if I could just ask the video directly? Content is already there — it's just not accessible in the way I need it.

The pain was real: educators scrolling through hour-long lectures to find specific topics, researchers hunting for data points in conference talks, and casual viewers trying to skip to the "good parts" of tutorials.

The Solution

I built VidWhiz — a Chrome extension that lets users chat directly with YouTube videos using natural language.

The architecture uses OpenAI's APIs and a RAG (retrieval augmented generation) system that: • Extracts video transcripts automatically • Chunks content intelligently by topic and context • Creates semantic embeddings for search • Serves real-time AI answers inside the browser

I focused on three core principles: speed (answers in <3 seconds), context-awareness (understanding follow-up questions), and clean UX (feels native to YouTube).

My Role

Product

I started by defining the problem based on my own frustration, then validated it through conversations with content creators, students, and researchers. I mapped the user journey to prioritize speed and ease — users needed answers without leaving YouTube or interrupting their workflow.

Backend

Built the core transcript processing service using Node.js and Azure Functions. The chunking logic splits transcripts by semantic meaning rather than arbitrary time stamps. Integrated OpenAI's embedding API for vector search and GPT-4 for response generation. Added caching layers to reduce API costs and improve response times.

Frontend

Developed the Chrome extension popup UI using React and Webpack. Created a clean, minimal interface that feels native to YouTube's design language. Implemented real-time chat functionality with typing indicators and context preservation across conversations.

DevOps

Deployed on Azure using serverless functions for scalability. Set up CI/CD with GitHub Actions for automatic testing and deployment. Implemented monitoring with Application Insights to track usage patterns and API performance.

Key Features

Chat with Any YouTube Video

Ask questions directly to videos in natural language

RAG-Powered Search

Semantic search over video transcripts for accurate context retrieval

Instant Answers

Get responses in under 3 seconds without leaving YouTube

Context Awareness

Follow-up questions understand previous conversation context

Lightweight Extension

Minimal footprint with seamless YouTube integration

Outcome

🚀 **Live with Growing User Base**: The extension is now live on Chrome Web Store with beta users actively using it daily.

⏰ **Time Savings**: Users report saving 60-90% of video watch time when looking for specific information.

🎯 **High Accuracy**: RAG implementation delivers contextually relevant answers with 85%+ user satisfaction.

📈 **Organic Growth**: Several founders and educators reached out to explore licensing the AI pipeline for their own applications.

💡 **Product Validation**: Clear demand for AI-powered content interaction tools, leading to discussions about expanding to other platforms.

Key Learnings

**User Education is Critical**: Many users initially assumed VidWhiz was "just ChatGPT with a skin." I learned that clear onboarding and specific examples dramatically improved adoption and trust.

**Context Window Management**: Long videos hit token limits quickly. I had to implement smart chunking strategies that preserved conversational context while staying within API constraints.

**Performance vs. Accuracy Trade-offs**: Balancing response speed with answer quality required careful prompt engineering and caching strategies.

Want to build an AI-powered tool like VidWhiz?

Let's Talk About Your Project