AI & Technical question

How will you design an algorithm for "continue watching segment" for a streaming services platform?

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

System design thinking for a recommendation and state-tracking feature, and how to measure whether it actually works.

Clarify this covers both resuming a specific episode and surfacing which show to continue in a row.
Define inputs: last watched timestamp per title, completion percent, recency, and device used.
Propose the simple core: store last playback position per user-title pair and resume within a few seconds of where they left off.
Rank the 'Continue Watching' row with recency-weighted decay, demoting or removing titles once fully completed.
Handle edge cases: multiple profiles per account, cross-device sync latency, and titles removed from the catalog.
Define success as resume accuracy and row click-through rate.

Separates storage and sync, an infra problem needing low-latency writes on pause or seek, from ranking, a product weighting problem.
Gives a concrete edge case most miss: a user 95% through should probably be marked watched, not continue.
Proposes a measurable quality proxy: percent of resumes that don't require rewinding or fast-forwarding.

Jumping to a complex ML model when a simple recency-plus-completion heuristic solves most of the problem.
Ignoring cross-device sync latency, a common real failure mode.

Chapters of the AI PM course that teach what this question tests.