~/portfolio·branch: main·v1.0.0
deployed · 2m ago
JKT · GMT+7·--:--:--
Back to work
/ 07voice-canvas·2025
Prototype · 2025

Voice Canvas.

An R&D experiment: a whiteboard you can talk to. Voice in → structured shapes out. Side-project that turned into a customer-facing demo.

Role
R&D · Stellar Labs
Client
Internal R&D
Year
2025
Duration
6 weeks
// previewVoice Canvas
prototype to demo6 wks
median sketch time90s
positive offsite feedback80%
follow-on product bets2
/ 01 — The problem

What was broken.

Could a designer narrate a wireframe into existence? Whisper had just released and we wanted to push the seam between voice and structured UI.

/ 02 — The approach

How I tackled it.

Six-week sprint. Used Whisper for transcription, a small Claude prompt to extract intent ('a circle inside a square, top-right'), TLDraw for the canvas substrate. Persisted voice metadata on each shape for later editing.

/ 03 — The outcome

What shipped.

Demoed at the company offsite — generated 80% positive feedback. Two product teams now exploring voice-input for their tools. Sketches that took 8 minutes in a normal whiteboard take 90 seconds.

/ Like what you see?

Let’s talk.