LaunchPad

Opening Screen
Star Map (Knowledge Graph) with stars, galaxies, solar systems (and little bitcamp alien ships!)
Using ElevenLabs and Gemini API to have voice and chat tutors that also use visualizations

Inspiration

We spend hours watching YouTube tutorials and lectures, but most of it doesn't stick. There's no feedback loop, no way to know if you actually understood something or just passively consumed it. We wanted to fix that by making learning from video feel more like a conversation than a one-way stream.

What it does

LaunchPad is a Chrome extension paired with a desktop app. While you watch a YouTube video, it pulls the transcript and generates quiz questions at key moments. Your answers get evaluated by AI and fed into a knowledge graph that visualizes your mastery as a 3D star map. Concepts you know well shine bright. Gaps in your knowledge show up as dim, unexplored stars.

How we built it

The extension intercepts YouTube playback and syncs transcript data to a local Express API. We used Gemini to generate context-aware questions from the transcript and evaluate free-text answers. The desktop app is built in React with Three.js powering the 3D constellation view. Everything runs locally with a file-backed knowledge graph that persists between sessions.

Challenges we ran into

Getting Gemini to produce consistent, structured JSON from open-ended transcripts took a lot of prompt iteration. The Three.js graph was also tricky to get right since we needed node positions, edge rendering, drill-down navigation, and smooth camera animations all working together without tanking performance.

Accomplishments that we're proud of

The 3D knowledge graph actually works and looks great. Watching your concept nodes light up as you answer questions correctly is genuinely satisfying. We also got the full pipeline running end to end: video to transcript to questions to evaluation to graph update, all in real time.

What we learned

Prompt engineering is harder than it looks. Getting an LLM to reliably return structured data with the right schema, handle edge cases gracefully, and stay on topic required way more iteration than we expected. We also learned a lot about building monorepos with shared TypeScript packages across a Chrome extension, an API, and a desktop app at the same time.

What's next

We want to add spaced repetition so LaunchPad resurfaces questions on concepts you're weak on over time. We'd also like to support more video platforms beyond YouTube and let users share their knowledge graphs with others.