Capstone projects by role
Put the course into practice with a guided capstone project for your role: what to build or produce, the steps, what "good" looks like, and a checklist so the result can be shown to a manager or employer as real evidence of your AI skills.
๐ฏ By the end of this lesson you will be able to:
- Choose a capstone project that fits your role and target AI role
- Plan and complete it in small steps using what you learned in the course
- Present the result as clear, credible evidence of your skills
In this lesson
Why a capstone matters
Reading about AI builds understanding. Doing something real with it builds skill, and gives you evidence: something you can show a manager, a hiring panel or a client. A good capstone is small enough to finish in two to four weeks of part-time effort, and realistic enough that it could be used at work.
Capstone projects by role
| Role path | Capstone project | What you produce |
|---|---|---|
| Developer โ AI Engineer | A RAG assistant over a public document set (e.g. a product's documentation), with citations, guardrails and an evaluation set | Working app, code repository, evaluation results, short design note |
| Tester โ AI Evaluation Engineer | An evaluation suite for an AI feature: 100+ cases across slices, a calibrated LLM judge, a red-team set and a CI gate | Dataset, rubric, judge calibration results, evaluation report |
| Tech Lead โ AI Architect | A reference architecture for an AI use case, with ADRs, model selection, governance controls and a cost model | Architecture diagram, 3โ5 ADRs, cost estimate, risk controls |
| PM / DM โ AI Delivery Manager | A delivery plan for an AI initiative: stages and gate criteria, evidence-based milestones, risk register and value case | Stage plan, gate criteria, risk register, ROI model |
| BA / PO โ AI BA / PO | A use case discovery and specification: scored opportunities, one-pager, AI requirements with examples and data readiness check | Opportunity scores, one-pager, requirement spec, readiness checklist |
| Scrum Master โ AI-ready Agile Coach | An AI adoption playbook for a team: working agreement, AI definition of done, ceremony helpers and a learning plan, trialled for two sprints | Playbook, trial results, retrospective findings |
Six steps for any capstone
- Pick a real problem you or your organisation recognise, and write it in two sentences with who has it and why it matters.
- Set success criteria up front: what will you measure, and what result would count as good? (Lessons 19 and 33)
- Build or produce the smallest useful version in about a week.
- Measure honestly with realistic examples, including failure cases, not just the best demo.
- Improve once or twice based on what the measurement showed, and record what changed.
- Write it up in one to two pages: problem, approach, results, limitations and what you'd do next.
What "good" looks like
| Weak capstone | Strong capstone |
|---|---|
| "I built a chatbot" | "A policy assistant that answers 86% of 120 test questions correctly with sources, and refuses when nothing relevant is found" |
| Only shows impressive examples | Shows typical cases, failures and what was done about them |
| No numbers | Baseline, target and measured results |
| Hides limitations | States limitations and risks clearly |
| Only the final result | Shows decisions and trade-offs along the way |
Capstone checklist
Before you call it done
โ Problem statement in two sentences โ Success criteria written before building โ Safe data only, and tools approved for it โ Realistic test set, including hard and failure cases โ Measured results with numbers โ At least one improvement cycle documented โ Limitations and risks stated โ One- to two-page write-up with a diagram โ Shared for feedback with at least one colleague or peer
This checklist also works as the structure of your write-up.
What this means for your role
Developer
Build the RAG assistant capstone; your evaluation results and design note matter as much as the code.
Tester / QA
Build the evaluation suite capstone; show judge calibration and slice results, which are rare and valued skills.
Business Analyst / PO
Do the discovery and specification capstone; a well-scored opportunity and a testable AI requirement are strong evidence.
Tech Lead / Architect
Do the reference architecture capstone; ADRs and a cost model show architect-level thinking.
Project / Delivery Manager
Do the delivery plan capstone; gate criteria, a risk register and a value case show you can deliver AI, not just talk about it.
Scrum Master / Agile
Run the team adoption playbook for two sprints; real retrospective findings make it credible.
Try it: plan your capstone today
- Choose the capstone for your role path, or adapt it to a real problem at work.
- Write the two-sentence problem statement and three success criteria with numbers.
- Put three one-week milestones in your calendar: build, measure and improve, write up.
Key takeaways
- A capstone turns knowledge into skill and gives you evidence others can see.
- Each role path has a capstone matched to its target AI role.
- Follow six steps: real problem, success criteria, small build, honest measurement, improvement, write-up.
- Strong capstones show numbers, failures, limitations and decisions, using only safe data.
Key terms
- Capstone project
- A final project that brings together what was learned in a course into one practical piece of work.
- Portfolio
- A collection of work samples that show your skills to others.
- Write-up
- A short document explaining a project's problem, approach, results and limitations.
Check your understanding
Answer all 5 questions. Score 3 or more to complete the lesson.
1. Which capstone description is strongest?
2. When should success criteria for a capstone be written?
3. What data should a personal capstone use?
4. Why include limitations in your write-up?
5. Which capstone fits a tester moving toward AI Evaluation Engineer?