CBA Certified Artificial Intelligence Professional

AI and Machine Learning Foundations20%

• Distinguish artificial intelligence, machine learning, deep learning and rule-based automation by how each produces its outputs.
• Classify a described problem as supervised, unsupervised or reinforcement learning and identify the target and features involved.
• Select an appropriate model family (linear models, tree-based models, clustering, neural networks) for a given problem and justify the choice by data type, interpretability need and scale.
• Interpret the roles of parameters, training, loss and optimisation in how a model learns, using simple numeric illustrations.
• Explain how core deep learning architectures (feedforward networks, convolutional networks, sequence models and transformers) differ in the data they suit.
• Identify the bias-variance trade-off, underfitting and overfitting from descriptions of model behaviour on training and unseen data.

Data Foundations for AI15%

• Assess the suitability of a dataset for a stated modelling task in terms of volume, representativeness, label quality and freshness.
• Select appropriate treatments for missing values, outliers, duplicates and inconsistent categories in a described dataset.
• Distinguish training, validation and test splits and identify splitting mistakes such as leakage and temporal contamination.
• Apply basic feature concepts, including encoding categorical variables, scaling numeric variables and simple feature engineering, to spreadsheet-scale examples.
• Evaluate the effect of class imbalance and sampling choices on what a model will learn.
• Identify data governance obligations relevant to AI datasets, including consent, minimisation and retention at concept level.

Model Development and Evaluation20%

• Calculate accuracy, precision, recall, F1 and error rates from a small confusion matrix or result table.
• Select the evaluation metric that matches a stated business cost of false positives versus false negatives.
• Interpret regression evaluation measures (mean absolute error, root mean squared error, R-squared) and compare candidate models with them.
• Distinguish the purposes of validation and test data and evaluate whether a described experiment supports its performance claim.
• Evaluate baseline comparisons, hyperparameter tuning descriptions and simple experiment designs for soundness.
• Interpret model explainability outputs, such as feature importance summaries, and identify their limits.

Deployment, Operations and the AI Lifecycle15%

• Evaluate whether a described business problem is a sound candidate for an AI solution and define measurable success criteria for it.
• Select an appropriate deployment pattern (batch scoring, real-time API, embedded on-device, human-in-the-loop) for a stated use case.
• Distinguish build, buy and API-based sourcing options for AI capability and assess their cost, control and data implications.
• Interpret monitoring signals to identify data drift, concept drift and performance degradation in a deployed model.
• Select appropriate responses to model degradation, including retraining, rollback and threshold adjustment.
• Identify the documentation and versioning practices that make an AI system auditable and reproducible.

AI Application Patterns and Generative AI15%

• Distinguish predictive, generative and retrieval-based AI applications by their inputs, outputs and failure modes.
• Explain how large language models produce text, including tokens, context windows and probabilistic generation, at concept level.
• Select prompt design techniques (role setting, examples, structured output, decomposition) appropriate to a described task.
• Evaluate when retrieval-augmented generation, fine-tuning or an unmodified foundation model is the right approach for a knowledge task.
• Identify hallucination, prompt injection and over-reliance risks in described generative AI workflows and select mitigations.
• Select appropriate human-in-the-loop checkpoints for AI-assisted work in fields such as drafting, translation, coding support and customer service.

Responsible AI, Governance and Risk15%

• Identify sources of unfair bias across the AI lifecycle, from historical data and labels to deployment context.
• Interpret simple fairness evidence, such as approval or error rates broken down by group, and identify the concern it raises.
• Distinguish transparency, explainability and accountability obligations for AI systems and match each to appropriate practices.
• Evaluate privacy and security risks specific to AI, including training-data exposure, model inversion at concept level, and misuse of generated content.
• Classify described AI systems into risk tiers of the kind used by the EU AI Act and identify the obligations that follow at concept level.
• Select appropriate governance controls for an AI deployment, including impact assessment, human oversight, documentation and incident response.