Product design question

Design a consumer feature using Project Astra (real-time multimodal assistant).

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

Product design for a genuinely novel interaction paradigm, real-time multimodal understanding, rather than a text chat feature.

Clarify the unique capability: Astra can see and hear the world continuously through a camera and microphone, understanding context in real time, unlike a typed query.
Identify a high-value use case: a visual assistant for everyday tasks, like identifying an object, translating a sign in real time, or guiding a repair step by step while watching the user's hands.
Design the interaction to be low-friction: always-listening within a session, responding proactively when it notices something relevant, rather than requiring a typed prompt for every question.
Address privacy directly: clear, visible indicators when the camera and microphone are actively processing, and easy one-tap pausing.
Define success: task completion rate for guided real-world tasks, and session length as a proxy for sustained, natural interaction rather than one-off queries.

Grounds the feature in Astra's actual unique capability, continuous real-time multimodal perception, rather than a text-based feature that happens to use a camera once.
Proposes a concrete, high-value use case, step-by-step guided repair or real-time translation, that clearly benefits from continuous visual context.
Designs proactive, low-friction interaction as the core UX, since typing to a camera-based assistant defeats its main advantage.
Addresses privacy with visible, simple controls as a first-class design requirement, not an afterthought, given the always-on nature of the capability.

Designing a feature that is really just visual search, underusing Astra's real-time, continuous understanding capability.
Ignoring privacy design for an always-listening, always-watching feature, which is a significant trust barrier.

How would you handle battery and performance constraints for continuous camera and microphone use?
What would you do if users found the proactive responses intrusive rather than helpful?