Design a Job Application Portal
Design versioned job publishing, searchable discovery, retry-safe applications, employer review, and applicant-visible status history.
Problem statement
For this exercise, assume verified employers create draft job listings, publish immutable listing versions, pause or close listings, and review applications. Candidates search active listings, read one exact listing version, and submit one application per candidate and listing with a versioned resume reference. An exact application retry returns the original application, while a materially changed retry conflicts. Employer decisions append to an ordered status history and never erase earlier candidate-visible states. Search may lag the listing authority briefly, but application admission rechecks the authoritative listing version and accepting state. Assume ten million active candidates, two million active listings, 100,000 search requests and 5,000 application submissions per second at peak, p95 search below 400 milliseconds, and accepted applications surviving one availability-zone failure. Resume editing, background checks, interview scheduling, messaging, billing, and recommendation ranking are outside scope. The implementation stack is prescribed: MongoDB, Express.js, React, and Node.js. Treat those technologies as fixed constraints and explain the data, API, server-process, and client boundaries needed to preserve the behavioral contract.
What to build
- Verified employers can publish, pause, and close versioned job listings with an auditable lifecycle.
- Candidates can search active listings and read results whose freshness and listing version are explicit.
- A candidate can submit one retry-safe application against an accepting listing version with a pinned resume reference.
- Employers can review applications and append authorized decisions that candidates observe as ordered status history.
At scale
- Ten million active candidates, two million active listings, 100,000 searches, and 5,000 application submissions per second are authored practice assumptions.
- Healthy search requests complete below 400 milliseconds at p95 with explicit projection freshness.
- Accepted applications survive one availability-zone failure and are never inferred solely from a search projection.