Eigenriver

home screen

Inspiration

I wanted a voice + gesture interface that felt like commanding the human fleet in Ender's Game!

What it does

Eigenriver is a game that lets you control your starfleet in real-time using natural voice commands and hand gestures. Think wake word → speech-to-intent → instant scene changes, camera moves, and object manipulation.

How I built it

Faster-Whisper for STT, GPT OSS 120B on Cerebras for structured output, Mediapipe for gestures, LangGraph for multi-turn command handling, and Three.js for the 3D layer.

Challenges I ran into

Making voice + hands feel natural, not clunky. Balancing latency vs accuracy. Getting large models to behave in a predictable, structured way.

Accomplishments that I'm proud of

A real-time, low-latency control loop that feels fun to use. Smooth hand tracking. The first iteration of a command grammar that could scale way beyond a hackathon demo.

What we learned

How to make multi-modal input actually feel good. The trade-offs between model size, inference speed, and UX. That structured outputs save you a ton of pain downstream.