Transcribe

Problem

A greenfield build: a text-transformation job pipeline. Clone the starter repo, use your own editor and AI tools, and build under a 60-minute timer.

The spec

A Job has a type (e.g., uppercase, dedupe, trim) and a payload. Build:

• A queue where jobs enter as queued
• A worker pool that processes up to N jobs concurrently
• A status lifecycle: queued → running → done, and failed → retried (with backoff) → dead-lettered after max attempts
• A way to inspect job status by id

Design decisions — the queue structure, the pool, the retry timing — are yours to make and defend.

What interviewers watch

• Do you model the job state machine before writing the worker?
• Is the concurrency limit real, or is the "pool" actually sequential?
• Are retries idempotent — no duplicate side effects when a job is retried?
• Do failing jobs get isolated, or does one failure stall the whole pool?

Submission

Export your code and AI chat transcript, note tools and time, and submit for a verdict.

Deliverables

• A job model with status lifecycle (queued → running → done | failed)
• A worker that processes jobs with bounded concurrency
• Retry with backoff and a dead-letter state after max attempts
• Tests for ordering, retries, and failure isolation