Assistive tech · Team ASTRA-X

UNSPOKEN

A chest-mounted wearable that helps visually impaired users sense obstacles and lets non-verbal users speak through gestures.

Award winner · Best Innovation Team

Source code ↗ CHAKRAVYUH post ↗

UNSPOKEN prototype on a table: a Raspberry Pi wired to two ultrasonic sensors and a breadboard

Problem

One affordable device for two groups: people who cannot see obstacles around them, and people who cannot speak.

Team build

Two ultrasonic sensors with vibration motors for left/right obstacle alerts, an MPU6050 IMU that turns lean gestures into spoken phrases ("Yes", "No", "Help", "SOS"), and a camera for scene description. A Python sensor-fusion loop on a Raspberry Pi Zero 2 W runs all of it, and a Node.js WebSocket server feeds a live React dashboard.

Hardest problem

A Pi Zero 2 W cannot run full TensorFlow or MediaPipe. The team used TensorFlow Lite models on the device and split the work: safety features run offline, while richer scene description uses the Gemini API when a connection exists.

Outcome

Won the Best Innovation Team award, with a ₹10,000 prize, at the 36-hour CHAKRAVYUH 1.0 National Hackathon, GITA Bhubaneswar, 30–31 January 2026.

Screenshots

Want to talk about this project?

I'm open to data internships, projects and hackathon teams.

Contact me See the full portfolio

More projects