# Design an AI meeting assistant: system design interview question

> Recorded audio to useful notes: long meetings, claims with evidence, access changes, and failures during costly model calls.

Source: https://sysround.com/questions/design-ai-meeting-assistant
Updated: 2026-11-05


Difficulty: hard. Levels: Senior, Staff+.

## How the interviewer opens it

"Design a service that turns recorded meetings into summaries, decisions, and action items. The service must preserve meeting access rules and base the notes on the meeting content."

## What to clarify first

- Input and scope
- Scale
- Data size
- Read:write mix
- Latency
- Evidence and uncertainty
- Access rules
- Consistency and availability
- Durability and retention
- Model privacy

## Scope of a strong answer

Core requirements:

- Accept recorded audio and report progress
- Create summaries, decisions, and action items
- Connect factual notes to transcript spans and audio timestamps
- Apply current meeting access rules to all artifacts

Usually out of scope:

- Live notes during a meeting
- Bots that join calls
- Calendar integrations
- Search across meetings
- Support for languages other than English

## What interviewers probe

### From uploaded audio to finished notes

Long audio files consume uneven compute. Failures can lose accepted work, repeat costly calls, or publish incomplete notes.

### What the notes can claim

Speech can be unclear, and later statements can reverse earlier ones. A fluent note with a timestamp can still make a false claim.

### Who can read meeting content

Notes copy private information into new artifacts. Access changes must affect every later server read, including requests for cached content.

### When meeting content must disappear

Workers can finish after expiry. Old backups can retain private content after the service blocks reads.

## Common mistakes

- The design treats a timestamp as proof that a claim is correct.
- The design copies an access list into notes and ignores later changes.
- The design assumes speaker labels prove identity.
- The design sends an entire three-hour transcript to one model call without checking limits.
- Retries repeat costly work without limits or publish duplicate results.
- The design blocks access after deletion but never removes old backup content.

## The bar by level

- **Senior:** Defines zone-safe acceptance, durable step boundaries, and conditional publication. Handles long meetings, revised decisions, provider failures, access removal, and expiry. Provides a complete deletion schedule.
- **Staff+:** Defines access and durability guarantees across caches, workers, failover, and restores. Proves the physical deletion deadline across all copies. Sets tenant capacity policies and model quality gates.

Practice this question with an AI interviewer: https://sysround.com/practice/new?problem=ai-meeting-assistant


---
SysRound (https://sysround.com) is an AI practice tool for system design interviews. It has a voice or typed AI interviewer and a shared whiteboard. At the end, you get feedback on each skill and a model answer. The AI feedback can be wrong.