Design a Fault-Tolerant Meeting Bot Allocation System
Quick Overview
Design the infrastructure between a web UI and a bot pool. Connect requirements and APIs to data modeling, consistency, scaling, failure recovery, observability, and the important design trade-offs.
Design a Fault-Tolerant Meeting Bot Allocation System
Company: Otter.Ai
Role: Software Engineer
Category: System Design
Difficulty: medium
Interview Round: Onsite
# Design a Fault-Tolerant Meeting Bot Allocation System
Design the infrastructure between a web UI and a bot pool. A user submits a meeting URL and optional credentials; the service returns acceptance asynchronously, atomically assigns an idle bot, and streams states through joined, failed, disconnected, or ended. Plan for 100,000 concurrent meetings and one million meetings daily.
### Constraints & Assumptions
- An acknowledged request cannot be silently lost.
- Messages may be duplicated and bots may fail mid-session.
### Clarifying Questions to Ask
- Are credentials stored or passed through?
- How quickly must state reach the browser?
### What a Strong Answer Covers
- Durable intake, queues, leases, retries, autoscaling, and state streaming.
- Idempotency, recovery, capacity exhaustion, and observability.
### Follow-up Questions
- How do you prevent double assignment?
- How do platform-specific adapters fit?
Quick Answer: Design the infrastructure between a web UI and a bot pool. Connect requirements and APIs to data modeling, consistency, scaling, failure recovery, observability, and the important design trade-offs.
Design a Fault-Tolerant Meeting Bot Allocation System
Otter.Ai
Dec 16, 2025, 12:00 AM
mediumSoftware EngineerOnsiteSystem Design
0
0
Design a Fault-Tolerant Meeting Bot Allocation System
Design the infrastructure between a web UI and a bot pool. A user submits a meeting URL and optional credentials; the service returns acceptance asynchronously, atomically assigns an idle bot, and streams states through joined, failed, disconnected, or ended. Plan for 100,000 concurrent meetings and one million meetings daily.
Constraints & Assumptions
An acknowledged request cannot be silently lost.
Messages may be duplicated and bots may fail mid-session.
Clarifying Questions to Ask Guidance
Are credentials stored or passed through?
How quickly must state reach the browser?
What a Strong Answer Covers Guidance
Durable intake, queues, leases, retries, autoscaling, and state streaming.
Idempotency, recovery, capacity exhaustion, and observability.