AI Engineering Specialization Project

Build one AI-enabled system around a real decision or workflow. Begin with a deterministic or statistical baseline, define the evaluation set before optimization, and preserve failed experiments.

Required Evidence

1. Dataset card with provenance, consent, exclusions, splits, and limitations.
2. Baseline implementation and comparison table.
3. Reproducible pipeline pinned to code, data, configuration, and model versions.
4. Evaluation harness covering quality, safety, robustness, latency, and cost.
5. Failure taxonomy with representative examples and severity.
6. Model/system card and NIST AI RMF risk register.
7. Prompt-injection or adversarial test report where applicable.
8. Human escalation, non-AI fallback, and rollback behavior.
9. Production telemetry and operating runbook.
10. Public-safe case study and 20-minute oral defense.

AI-Use Declaration

Record every model used to create code, data, tests, or prose; what was accepted or rejected; and how important claims were independently verified.