Back to course overview
Module 7Capstone: production prompt 15 min

Presentation

Present the system, defend the design, review a peer's spec — and claim the Prompt Engineering Specialist credential.

Finally, present your system, survive two rounds of questions, and review a peer's spec with the same rigor you applied to your own. Then the credential.

The 8-minute design review (not a demo)

  1. 1The task and the stakes (1 min) — inputs, consumers, and the two worst cells of your failure-cost table. Stakes first: they justify everything after.
  2. 2The contract (1 min) — schema on screen; point at the null policy and one enum, and name who consumes each.
  3. 3One case, end to end (2 min) — a boundary case: input → reasoning → JSON → validators → route. Then the same case's line in the golden set. This single walkthrough proves the whole architecture.
  4. 4The evidence (2 min) — golden-set composition, current scores, and your A/B log line: hypothesis, flips, decision. Show the regression you caught or the variant you rejected — rejected changes are the best proof the method works.
  5. 5Limits and the playbook (2 min) — known limits verbatim from the spec; then 'when it breaks, here's who does what'. End on operational honesty, not a sales note.

Questions you should expect (because reviewers are instructed to ask them)

  • 'What happens when someone pastes an instruction into your input?' — demo or cite your injection case.
  • 'Why should I believe the eval isn't overfit?' — holdouts, production-refresh policy.
  • 'What breaks first at 10× volume?' — cost budget, circuit breaker, human-queue capacity.
  • 'Show me the last change you decided NOT to ship.' — your experiment log answers this in one line, which is exactly why it exists.

Peer review & credential

  1. 1Review one peer's full spec with the doc test from last lesson, plus rubric scores: contract quality / example & CoT judgment / grounding integrity / eval rigor / guardrail completeness (20% each).
  2. 2Apply the one-change rule from Foundations: incorporate your reviewer's strongest point before final submission, logged as a version bump.
  3. 3Submit: requirements brief, frozen system artifacts, prompt spec, presentation deck or script, your peer review.

Passing earns Prompt Engineering Specialist (credential format EDOVA-PE-2026-XXXX) — certifying that you design structured, grounded, evaluated, guarded prompt systems and document them to a professional standard. It's the certificate previewed on our Academy page; now you know exactly how much system sits behind it.

Where to go next

All four tracks continue from here. Building RAG Applications takes Module 4's overview to production retrieval. Agentic AI Systems turns your schema'd, guarded prompts into multi-step tool-using agents — it follows Building RAG Applications, so take RAG first. Analytics-track learners: the Conversational Analytics Agent course applies everything here to governed data. Your triage assistant and your capstone spec travel with you into all of them.