1Z0-1122-25 Syllabus — Learning Objectives by Topic

Use this syllabus as your source of truth for 1Z0‑1122‑25.

What’s covered

• Topic 1: AI & ML Fundamentals
• Topic 2: Data, Features, and Quality
• Topic 3: Training and Evaluation Basics
• Topic 4: Generative AI Foundations
• Topic 5: Responsible AI, Security, and Governance
• Topic 6: OCI AI Services and Reference Architectures

Topic 1: AI & ML Fundamentals

Practice this topic →

1.1 AI vs ML vs deep learning and common use cases

• Differentiate artificial intelligence, machine learning, and deep learning in practical terms.
• Classify common problems as classification, regression, clustering, recommendation, or anomaly detection.
• Given a scenario, choose ML vs rules-based automation vs traditional analytics.
• Explain what training data is and why labeled data changes the type of solution you can build.

1.2 The ML lifecycle (from problem framing to operations)

• Describe the typical ML lifecycle: define, collect, prepare, train, evaluate, deploy, monitor, iterate.
• Explain why baseline models and clear success metrics are required before optimizing.
• Recognize the difference between experimentation code and production ML systems (MLOps mindset).
• Identify where bias, security, and governance requirements must be addressed in the lifecycle.

1.3 Data-driven decision making and model limitations

• Explain why correlation does not imply causation and how spurious features can mislead models.
• Recognize common failure modes: distribution shift, overfitting, underfitting, and concept drift.
• Given a scenario, identify whether more data, better labels, or better features is the primary need.
• Explain the difference between model accuracy and business impact.

Topic 2: Data, Features, and Quality

Practice this topic →

2.1 Data types, labeling, and dataset splits

• Distinguish structured, semi-structured, and unstructured data and how they are commonly used in ML.
• Explain why train/validation/test splits exist and what each split is used for.
• Identify label quality issues (noise, ambiguity) and their effect on model performance.
• Given a scenario, choose a split strategy that avoids leakage (for example, time-based splits for time series).

2.2 Feature engineering fundamentals

• Explain the goal of feature engineering and how it can improve simple models.
• Recognize common transformations: normalization/standardization, one-hot encoding, and binning.
• Given a scenario, choose an approach for missing values (impute, drop, sentinel values).
• Explain why feature selection can reduce overfitting and improve interpretability.

2.3 Data quality, imbalance, and drift

• Identify common data quality issues: duplicates, outliers, inconsistent formats, and missingness.
• Explain why class imbalance can break accuracy as a useful metric and what to do instead.
• Recognize symptoms of data drift and concept drift in production.
• Given a scenario, choose monitoring signals that detect data quality problems early.

Topic 3: Training and Evaluation Basics

Practice this topic →

3.1 Training workflows and validation

• Explain the purpose of validation during training and why test data must remain untouched.
• Recognize when cross-validation is appropriate and when it is not (for example, time series).
• Describe the difference between hyperparameters and model parameters at a conceptual level.
• Given learning curves, identify overfitting vs underfitting and the next best action.

3.2 Metrics for common ML tasks

• Choose appropriate metrics for classification (precision, recall, F1, ROC-AUC) based on the cost of errors.
• Choose appropriate metrics for regression (MAE, MSE/RMSE, R-squared) based on sensitivity to outliers.
• Interpret a confusion matrix and identify common trade-offs when adjusting thresholds.
• Explain why accuracy can be misleading and when to prefer calibrated probabilities.

3.3 Error analysis and iteration

• Explain the purpose of error analysis (finding systematic failure patterns).
• Given a scenario, choose whether to collect more data, relabel data, or engineer features.
• Describe what a baseline model is and why it is valuable even if it is simple.
• Recognize common sources of evaluation leakage (using target-derived features, peeking at test data).

Topic 4: Generative AI Foundations

Practice this topic →

4.1 LLM concepts (tokens, embeddings, context)

• Explain what tokens and context windows are and how they affect cost and answer quality.
• Describe embeddings and why they enable semantic search and retrieval.
• Differentiate base models, instruction-tuned models, and chat models at a conceptual level.
• Given a scenario, choose when summarization, extraction, or classification via an LLM is appropriate.

4.2 Prompting basics and failure modes

• Structure a prompt using role/context, task, constraints, and output format.
• Recognize common prompt failure modes: ambiguity, missing constraints, and hallucinated details.
• Explain how temperature/top-p affect variability and when determinism is preferred.
• Given a scenario, choose few-shot examples vs explicit rules vs tool use to improve reliability.

4.3 RAG at a high level (grounding and citations)

• Explain the purpose of retrieval-augmented generation (ground answers in trusted data).
• Identify the main RAG stages: ingest, chunk, embed, index, retrieve, generate, cite.
• Recognize why chunking and metadata impact retrieval quality.
• Given a scenario, choose RAG instead of fine-tuning to keep knowledge up to date.

Topic 5: Responsible AI, Security, and Governance

Practice this topic →

5.1 Responsible AI principles and risk categories

• Describe responsible AI principles: fairness, accountability, transparency, and safety.
• Identify risks across the lifecycle: biased data, misuse, unsafe outputs, and privacy violations.
• Explain why documentation (model cards, dataset notes) supports auditability and trust.
• Given a scenario, choose appropriate human-in-the-loop controls for high-impact decisions.

5.2 Security and privacy basics for AI systems

• Explain why least privilege and access boundaries apply to training data and prompts.
• Recognize prompt injection and data exfiltration as LLM-specific security threats.
• Choose protections for sensitive data: encryption, redaction, access controls, and logging hygiene.
• Given a scenario, identify where PII can leak (inputs, outputs, logs, caches) and mitigations.

5.3 Monitoring, governance, and change control

• Identify monitoring needs: quality metrics, drift, latency, cost, and safety signals.
• Explain why model versioning and approval workflows reduce production risk.
• Recognize when to retrain vs adjust features vs roll back to a prior model version.
• Given a scenario, choose governance controls for regulated workloads (audit trails, segregation of duties).

Topic 6: OCI AI Services and Reference Architectures

Practice this topic →

6.1 OCI Data Science essentials

• Identify core OCI Data Science concepts: projects, notebooks, jobs, models, and model deployments.
• Describe how training artifacts and datasets are commonly stored (for example, Object Storage).
• Recognize when to use managed jobs vs interactive notebooks for training workflows.
• Explain how model deployments expose endpoints and require IAM-controlled access.

6.2 OCI AI Services essentials (Language, Speech, Vision)

• Describe the purpose of OCI AI Services and how they differ from training custom models.
• Given a scenario, choose Language vs Vision vs Speech services based on the input and desired output.
• Recognize typical integration patterns: REST APIs, SDKs, and event-driven processing.
• Explain key operational considerations: quotas, rate limits, and cost/latency trade-offs.

6.3 Secure, cost-aware AI architecture on OCI

• Identify the OCI services commonly used around AI workloads: IAM, Vault, networking, logging, monitoring, Object Storage.
• Given a scenario, choose public vs private network access patterns and justify security trade-offs.
• Explain how compartments, tagging, and budgets support governance and cost tracking.
• Recognize basic reliability practices: multi-AD design, retries/backoff, and idempotent processing.