Capstone project
Last reviewed: 2026-06-28 — see the freshness policy.
Learning objectives
After this chapter you will be able to:
- Integrate everything in this bootcamp into one defensible architecture.
- Produce the full SA deliverable for a realistic, multi-jurisdiction brief.
- Defend your design against a rubric covering correctness, compliance, cost, and clarity.
The brief
You are the Solution Architect for a growing digital-health company. It runs a patient-facing app and a clinician portal, integrates with hospital EHRs, and wants to use its data for AI features and real-world-evidence partnerships. It operates in the United States and Canada and is expanding into the EU. Design the end-to-end platform.
The brief deliberately spans the whole bootcamp: clinical data + interoperability, multi-jurisdiction compliance, a cloud/hybrid platform, a data/RWD layer, an AI feature, and the SA craft to tie it together and defend it.
What to deliver
Produce the SA deliverable set (the structure from Part 9):
- Requirements summary — problem, scope, stakeholders, FRs, and measurable NFRs (use the discovery template).
- Context & container diagrams — C4 L1 and L2.
- Key ADRs — at least three significant, hard-to-reverse decisions with trade-offs.
- Compliance mapping — controls mapped to HIPAA, PIPEDA/PHIPA, and GDPR/EHDS with data-residency per jurisdiction (see regional compliance).
- Cost model — a back-of-envelope TCO for build and run.
- Security review pack — threat model + control checklist (security review).
- Stakeholder brief — one-paragraph framings for a clinician, a CISO, and a CFO (stakeholders).
Design questions you must answer
- Interoperability: how do you ingest from EHRs (HL7v2/FHIR) and expose patient/clinician access (SMART)?
- Jurisdiction & residency: where does each population's data live, and how do you prevent hidden cross-border hops (logs, backups, model APIs)?
- Platform: which cloud(s), and where does on-prem/hybrid fit (capability map, on-prem & hybrid)?
- Data & RWE: how do you model the data (OMOP) and enable a partnership without moving PHI (tokenization / clean rooms)?
- AI feature: pick one (clinical RAG, an agent, or imaging/NLP) and keep PHI in-boundary; if it informs care, address SaMD.
- The SA call: name your dominant quality attribute and the biggest trade-off you made, and defend it.
Rubric
| Dimension | What "good" looks like |
|---|---|
| Correctness | The design actually solves the brief; data flows make sense end to end |
| Compliance | All three jurisdictions addressed; residency handled; controls mapped, not hand-waved |
| Trade-offs | Decisions are explicit, with alternatives considered (real ADRs) |
| Cost | A credible TCO with the main HLS cost drivers identified |
| Communication | A clinician, CISO, and CFO could each understand their part |
You "graduate" by passing this review (see Bootcamp format). There is no single right answer — there is a defensible one, and the defense is the point.
Draft your own submission before reading the capstone exemplar. It's a complete worked example against this exact brief and rubric — useful for calibration, not for copying.
Check yourself
- What is your dominant quality attribute for this brief, and what did you trade for it?
- For each of the US, Canada, and EU, where does that population's data reside and why?
- Which one AI feature did you choose, and how does it keep PHI in-boundary?
Further reading
- Reference architectures — worked exemplars
- Discovery & requirements · Security review