Product design question

Design Google Glass for the blind people.

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Whether you can repurpose an existing hardware platform for a genuinely different sensory need rather than a cosmetic tweak.

Clarify scope: Google Glass reimagined as an audio and haptic-first assistive device for blind and low-vision users, not a visual display product.
Identify the core need: real-time environmental awareness, like reading signs, recognizing faces, and identifying obstacles, that sighted people get visually.
Prioritize a camera-plus-AI description pipeline that narrates the environment through bone-conduction audio, keeping ears free to hear ambient sound.
Address a key risk: audio overload, since constant narration would overwhelm the user, so design for on-demand and priority-based alerts only.
Propose a concrete flow: a tap or voice cue triggers a scene description, while obstacle alerts trigger automatically and urgently.
Define success metrics: successful navigation and object-identification task completion, with a guardrail on false or missed obstacle alerts.

Reimagines the hardware's output entirely, audio and haptic instead of a visual display, matching the actual user's sensory needs.
Prioritizes on-demand narration over constant description, addressing the real risk of overwhelming the user with audio.
Separates automatic urgent alerts, like obstacles, from on-demand information, like reading a sign, showing thoughtful interaction design.
Names missed obstacle alerts as a safety-critical guardrail, since a failure here has real physical consequences.

Keeping a visual display as the primary interface, ignoring that the target user can't see it.
Designing constant audio narration without addressing the real risk of overwhelming or distracting the user.

How would you test the accuracy of the AI's scene descriptions before trusting it for safety-critical alerts?
What would you do for identifying people, given the privacy sensitivity of facial recognition?