Know exactly what the exam asks, rehearse the timed practical, test yourself with a twenty-question mock, and claim your certificate.
Module 14 ~60 min read + lab No codePrerequisites: Module 13: Capstone submitted, and every earlier module marked complete.
The final assessment has two parts, taken in one sitting of about two hours and twenty minutes including a break.
| Part | Format | Time | Weight |
|---|---|---|---|
| Theory | 40 multiple-choice questions covering all modules, one correct answer each, no negative marking | 60 minutes | 60% |
| Practical | A live, timed exercise: map a given process and build it as a no-code automation with a human approval step | 60 minutes | 40% |
The theory questions are in the same style as the knowledge checks at the end of each module. Roughly: workflow mapping 6 questions, agents and prompting 8, automation and AI steps 8, connectors and MCP 4, human-in-the-loop 6, security and governance 4, monitoring 4.
You will receive a one-paragraph process description you have not seen before. It will always have the same shape: a trigger, a piece of text a person currently reads, a decision, a record to update, and a message to send. Here is a sample of the kind you will get:
Facilities receives maintenance requests by email. A coordinator reads each one, decides whether it is urgent (safety or outage) or routine, adds a row to the maintenance tracker with the location and category, and replies to the requester with an expected response time (same day for urgent, five working days for routine). Urgent requests are also forwarded to the on-call engineer.
In 60 minutes you must produce:
| Criterion | Marks |
|---|---|
| Map is readable, steps tagged, tools tiered | 8 |
| Flow runs end to end on the examiner's test input | 10 |
| AI step produces the needed fields reliably (structured output, category and urgency) | 6 |
| Approval gate shows the full action; timeout is safe; reject path works | 10 |
| Audit-log row and the five-line note are complete | 6 |
A flow that sends the external message without a gate scores zero on the gate criterion and cannot exceed 25 of 40. Partial builds earn partial marks: a map plus a working AI step with no reply yet is worth more than a complete flow with no map.
Most lost marks in the practical are lost in the first ten minutes to logins and missing templates. Do this the day before.
In the instructor-led track, the certificate is issued jointly by IEEE Computer Society Region 8 and the AI Caravan programme once you have met the attendance requirement, passed the exam and presented the capstone. It states the track name, the 28 hours, and the three competency areas: workflow mapping, agents and automation, and controls with human-in-the-loop design.
On this site, your progress is tracked in your browser. When every module is marked complete, the certificate page unlocks a self-study completion certificate you can download. It is not the IEEE certificate, but it records that you finished the material and the labs.
The best next step is to run your capstone for real for a month, with the dashboard and the changelog, and then present the numbers to whoever owns the process. That first month of evidence is worth more than any further course. When you are ready for the developer side of the same ideas, the LangGraph tutorial and the MCP guide on this site pick up exactly where this course stops.
A mock practical under exam conditions. Use the sample brief in Section 2, or ask a colleague to write you a fresh one-paragraph process in the same shape. Set a 60-minute timer and do not pause it.
Work through the exam-day checklist above until every box is ticked. Open the templates in tabs.
Sketch the future-state swimlane for the brief. Tag each step and tier each tool. Stop at ten minutes even if it is rough.
Trigger, AI extraction step with structured output, tracker row, drafted reply. Test with one input until it runs.
Approval on the reply and any forward, full content visible, 24-hour safe timeout, reject path to a "needs human" label, one audit-log row.
Write the five lines: data classes, highest tier, failure behaviour, kill switch, one risk to watch.
Score your work against the 40-mark criteria honestly. Anything under 28 tells you which module to revisit before the real exam. Then take the mock exam below and aim for 14 of 20 or better.
Your mock practical (map, workflow export, approval screenshot, audit row, note) with your self-assessed score, and your mock exam score. Save as M14-mock.
Pick one answer per question, then check your score. These mirror the style of the final exam.
Answer in your own words first, then open the model answer.
Forty multiple-choice questions in 60 minutes worth 60%, then a 60-minute practical worth 40%. Pass is 70% overall. Attendance of 80% and a presented capstone are also required.
Sending the external message without an approval gate, or with a gate that shows no content. Use the Module 10 send-and-wait pattern with the full draft in the message and a safe timeout.
Ten minutes to map and tier, thirty to build the happy path with a structured AI step, fifteen for the gate and audit row, five for the note. Stop each phase on time.
Further reading: Your course certificate · LangGraph tutorial · MCP guide
