AI System Design Interview Questions
Dashboard
0%
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
51 small wins to finish your pathNext question →

Q20IntermediateSystem design

Design a meeting summarisation system: audio → transcript → summary → action items.

30-second answerSay your answer out loud first, then reveal.
Meeting summarisation pipeline: audio goes to streaming speech-to-text for live captions and to object storage for batch speech-to-text with diarisation, then cleaning, topic segmentation, an LLM per segment, an LLM reduce step producing the summary, decisions and action items, a review UI, and on approval a push to Jira, Asana or email.

Deep dives

  1. STT quality: accents, code-mixed Hindi/English, domain jargon. Use a custom vocabulary or glossary (product names, people), and evaluate word error rate per language.
  2. Diarisation: "who said what" matters for action-item owners. Map speakers to calendar attendees, and let users fix labels.
  3. Long meetings: a 1-hour meeting is ~10K words ≈ 13K+ tokens. It may fit in context, but hierarchical summarisation improves structure and allows parallel processing.
  4. Faithfulness: every action item links to a timestamp. Instruct the model not to invent owners or dates ("owner: unassigned" if unclear).
  5. Privacy and consent: recording notices, retention policies, access limited to attendees, redaction of sensitive segments, data residency.
  6. Latency: summary within ~1–2 minutes after the meeting ends is usually acceptable, so batch processing is fine.

Metrics: action-item precision and recall (vs human notes), user edit rate, ratings, adoption.

PreviousDesign an AI reply suggestion feature (smart replies) for a customer-support inbox used by 5,000 agents.

Every expert started right here.