AI & Technical question

Design a video chat infrastructure. How do you handle the scalability of the system with the least amount of cost?

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

Systems design trading off scalability and cost for real-time media infrastructure, with a named architecture pattern.

Clarify scale: concurrent call volume and typical group size, since architecture differs significantly between 1:1 and large group calls.
Propose the core approach: a Selective Forwarding Unit architecture rather than full mesh peer-to-peer, since mesh doesn't scale past a few participants.
Address cost: use adaptive bitrate and simulcast so the server forwards only the necessary quality per participant's bandwidth, reducing costs at scale.
Address regional cost and latency trade-off: deploy media servers in regional points of presence to reduce cross-region bandwidth cost and latency.
Add elasticity: auto-scale media server capacity based on real-time concurrent call load rather than provisioning permanently for peak.
Define success as cost per concurrent-minute of video and call quality metrics, packet loss and latency, under load.

Names the correct real-world architecture pattern, SFU over full mesh or MCU, with clear reasoning, showing genuine systems knowledge.
Explicitly ties cost reduction to a concrete mechanism, simulcast and regional scaling, not just 'use the cloud.'
Balances scalability and cost with a measurable metric, cost per concurrent-minute, rather than treating them as unconnected goals.

Proposing full mesh peer-to-peer for group calls, which doesn't scale past a handful of participants.
Ignoring cost entirely and only discussing scalability, missing half the question.

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