Skip to lesson
Exit
The Interview Loop1 / 2

2 min lesson

The practical build round

Put yourself in this case: "In the multi-hour build round, why does a small working usage-to-invoice slice beat an ambitious revenue-subledger architecture you didn't finish?" Give the clearest next step.

Step 1 of 2

The signature stage: a multi-hour, possibly full-day build where you design a finance-systems solution to a deliberately ambiguous prompt and implement enough to prove the system design. Full AI access is expected and you're evaluated on product sense, autonomy and judgment - not on a polished, finished artifact.

A functional understanding with clear trade-offs beats a polished half-feature. The round simulates the job: scope an ambiguous finance problem, ship a demoable slice and defend the architecture you chose against the ones you didn't.

The prompt is usually one of a few recognizable shapes; scope ruthlessly to one and take it end to end:

  • A usage-to-invoice pipeline: meter → rate → invoice.
  • A reconciliation or anomaly-detection service.
  • A revenue-subledger sketch feeding the GL.
  1. 1Orient and disambiguate first. Read the prompt for what's deliberately vague, then ask one sharp clarifying question and state your assumptions about grain, scale and audit needs.
  2. 2Cut to a demoable slice fast. Pick the one path that proves the system works end to end - for a usage-to-invoice prompt, that's one meter, one rating rule, one invoice - and protect it from creep.
  3. 3Design for idempotency and reconciliation up front. A re-run must not double-bill and you should be able to reconcile output against an immutable source. Say this aloud as you build.
  4. 4Build correct increments with seams. Leave clear boundaries where the GL posting, multi-currency or SOXSarbanes-Oxley Act. A US law that forces companies to keep auditable controls over any system that affects their financial reporting. Press Enter for the full definition. controls would slot in later, so the design reads as extensible.
  5. 5Mark where agents fit, with guardrails. Name one place an AI agent could safely automate (anomaly flagging, draft reconciliations) and the guardrail that keeps it from touching the ledger unreviewed.
  6. 6Budget time for a clean closeout. Reserve the last block to present what you built, what's next and the risks you deferred - the closeout is graded.