Assessment framework
Continuous assessment, five stage gates and a defended capstone. The pass standard mirrors the qualification pack: 70% aggregate and 70% on every technical NOS independently.
Instruments
| Instrument | Frequency | Format | Weight |
|---|---|---|---|
| Module quiz | Per module | Auto-graded objective, concept-bank driven | 10% |
| Coding challenge | Per module | Auto-graded, hidden test cases, time-boxed | 10% |
| Lab portfolio | Continuous | Rubric-graded artefacts in version control | 15% |
| Stage gate | 5 gates | Theory paper + extended practical | 25% |
| Workplace simulation | 8 scenarios | Observed performance against behavioural rubric | 10% |
| Capstone | Once | Project + report + panel defence | 25% |
| Professional conduct | Continuous | Punctuality, collaboration, code review, communication | 5% |
Stage gates
| Stage | Weeks | Gate |
|---|---|---|
| A Foundations | 1–6 | Gate 1 — Foundations examination |
| B Core AI | 7–12 | Gate 2 — NOS 8121 / 8122 practical |
| C Applied GenAI | 13–18 | Gate 3 — RAG/agent system with evaluations |
| D Specialisation | 19–22 | Gate 4 — deployed track project |
| E Capstone & OJT | 23–24 | Gate 5 — capstone defence |
| P Professional (parallel) | 1–24 | Continuous evaluation |
Progression rules
- Every stage gate must be passed at ≥70%; two re-attempts permitted with mandatory remediation between attempts.
- Statutory subjects C02 (SSC/N8121) and C05 (SSC/N8122) must each be passed at ≥70% independently — they cannot be compensated by other marks.
- Certification requires ≥70% aggregate, a completed capstone, and 90 hours of OJT with supervisor sign-off.
- Attendance minimum 80% of scheduled practical hours.
- Plagiarism or unreproducible results in any graded artefact voids that assessment.
Capstone rubric
| Criterion | Marks | Evidence required |
|---|---|---|
| Problem framing & requirements | 10 | SRS with functional and non-functional requirements; measurable success criteria |
| Design quality | 15 | HLD/LLD with rationale, assumptions, constraints, dependencies, interfaces |
| Implementation correctness | 20 | Working system; tests passing; coverage evidence |
| Evaluation rigour | 20 | Pre-declared metrics; baseline comparison; retrieval and generation separated where applicable; honest error analysis |
| Engineering & operations | 15 | Version control discipline, CI, containerisation, observability, runbook |
| Safety, security & governance | 10 | Guardrails, PII handling, threat consideration, documented limitations |
| Communication & defence | 10 | Report clarity, demonstration, quality of answers under questioning |
Integrity rule: unverifiable or unreproducible results score zero on evaluation rigour. Every reported number must be reproducible from the repository by an assessor.