Phase 0Weeks 1–2
Foundation & ways of working
A reviewed pull request and an evidence-aware working rhythm.
Human-confirmed gate
Merged PR, green checks, individual walkthrough and baseline rubric.
Program command center
Build engineers who can prove what they ship.
A sixteen-week pathway from disciplined delivery to bounded agentic systems—using verified artifacts, transparent rubrics and human-owned phase gates.
Program length16 weeks80 focused build days
Delivery model20 / 80Theory / applied work
Cohort design12–20Mentored, remote-enabled
Decision modelHuman-ownedAI cannot issue a pass
The learning arc
One evidence spine, five accountable phases
- 0FoundationWeeks 1–2
- 1PlatformWeeks 3–5
- 2DataWeeks 6–8
- 3AgenticWeeks 9–12
- 4CapstoneWeeks 13–16
Phase 1Weeks 3–5
Cloud-native delivery
A containerized service delivered through CI with useful telemetry.
Human-confirmed gate
Authenticated service passes positive and negative probes in an isolated environment.
Phase 2Weeks 6–8
Data & context engineering
A reproducible RAG service over a governed real-world corpus.
Human-confirmed gate
Held-out evaluation demonstrates retrieval quality, grounded answers, refusal and cost/latency bounds.
Phase 3Weeks 9–12
AI & agentic engineering
A bounded tool-using system with state, evaluations and layered safety controls.
Human-confirmed gate
Golden tasks and adversarial suites pass; unsafe actions fail closed and require human approval.
Phase 4Weeks 13–16
Real-world capstone
A useful ALANS product increment with individual evidence and operational ownership.
Human-confirmed gate
Deployed increment, evidence bundle, individual defence, product-owner review and human panel decision.
The certification contract
Evidence is necessary. Judgment stays accountable.
A commit, deployment or evaluation is an artifact—not automatically proof of competence. Verifiers establish what happened; approved rubrics define what it means; named reviewers own the decision.
- Every gate names its required evidence and rubric version.
- Safety-critical criteria cannot be averaged away.
- Missing evidence produces an abstention, never an invented score.
- Certification remains separate from employment selection.
Assessment dimensions
Observable behaviours, not personality proxies
01
Technical capability
Correctness, security, testability, operability and quality of the demonstrated work.
02
Problem solving
Problem framing, evidence-based choices, debugging and learning from failed approaches.
03
Collaboration & communication
Review quality, shared context, respectful coordination and clear technical explanation.
04
Ownership & reliability
Transparent commitments, safe follow-through, risk escalation and operational care.
05
Responsible AI practice
Effective AI use with verification, privacy, safety controls and preserved human understanding.
Manual-first pilot
Automation earns authority; it does not begin with it.
The first cohort establishes human baselines. Later assessment assistance runs in shadow mode and is measured for agreement, abstention, overrides and evidence coverage before any governed recommendation is enabled.