AI-300 Exam Training Course Study Plan
AI-300: Microsoft Certified Machine Learning Operations MLOps Engineer Associate Training Course
The AI-300 training course is designed for professionals preparing for Microsoft Certified: Machine Learning Operations (MLOps) Engineer Associate (beta). It focuses on operationalizing machine learning and generative AI solutions on Azure by connecting Azure Machine Learning, Microsoft Foundry, GitHub Actions, Bicep, Azure CLI, evaluation workflows, observability, RAG optimization, and model customization.
This 5-week study plan is built as a first-attempt preparation path for learners who need systematic mastery of MLOps and GenAIOps execution logic. The course uses official AI-300 blueprint domains as weekly anchors and uses the H3 decomposed knowledge points as daily learning targets.
What This Plan Offers
Comprehensive coverage of the five official AI-300 domains:
Design and implement an MLOps infrastructure
Implement machine learning model lifecycle and operations
Design and implement a GenAIOps infrastructure
Implement generative AI quality assurance and observability
Optimize generative AI systems and model performance
Daily goals and tasks based on decomposed H3 knowledge points.
Pomodoro Technique and Forgetting Curve Principle for repeatable retention.
Practice-oriented outputs including command drills, YAML checklists, scenario cards, trace maps, evaluation tables, endpoint troubleshooting notes, and RAG tuning comparison sheets.
Who This Plan Is For
Data scientists moving from experimentation into production model operations.
DevOps engineers supporting Azure Machine Learning, Microsoft Foundry, GitHub Actions, Bicep, and CLI-based automation.
AI engineers who deploy, evaluate, monitor, and optimize generative AI applications and agents.
Structured learners who need daily tasks, scenario practice, and exam-question transformation drills.
Final Outcome
By the end of this training course, the learner should be able to map each AI-300 scenario to a domain, identify the missing operational dependency, select the correct Azure Machine Learning or Microsoft Foundry action, reject plausible distractors, and verify the result through commands, metrics, logs, traces, or evaluation output.
Week 1 Study Plan - Design and implement an MLOps infrastructure
Primary Objective for the Week:
Master the Design and implement an MLOps infrastructure domain by studying its decomposed H3 knowledge points as operational workflows. By the end of the week, the learner should explain what to configure, why the configuration is required, how the exam turns it into a scenario, and which verification signal proves the answer.
Learning Methodology for the Week:
Daily duration: 3.5 to 4 hours, organized as 6 Pomodoro blocks.
Pomodoros 1-2: Knowledge Explanation reading and why-layer extraction.
Pomodoros 3-4: Diagram, command, YAML, workflow, or scenario artifact creation.
Pomodoro 5: Flashcard creation and wrong-option analysis.
Pomodoro 6: Forgetting-curve review of previous days and mixed scenario questions.
Day 1 - Create and manage resources in an Azure Machine Learning workspace
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Create and manage resources in an Azure Machine Learning workspace.
Tasks:
Read the Knowledge Explanation section for Create and manage resources in an Azure Machine Learning workspace and extract Machine Learning workspace, Workspace managed identity, Datastore, Compute cluster plus the failure trigger: Training jobs queue or fail when compute quota, datastore identity, private DNS, or storage firewall dependencies are missing.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 2 - Create and manage assets in an Azure Machine Learning workspace
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Create and manage assets in an Azure Machine Learning workspace.
Tasks:
Read the Knowledge Explanation section for Create and manage assets in an Azure Machine Learning workspace and extract Data asset, Environment, Component, Registry plus the failure trigger: Pipelines fail when an asset version is omitted, archived, or resolved from the wrong workspace scope.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 3 - Implement infrastructure as code for Azure Machine Learning
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Implement infrastructure as code for Azure Machine Learning.
Tasks:
Read the Knowledge Explanation section for Implement infrastructure as code for Azure Machine Learning and extract Bicep module, GitHub Actions workflow, Federated credential, Azure CLI deployment plus the failure trigger: Automation fails before resource deployment when the workflow lacks id-token permission or the federated credential subject does not match the branch.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 4 - Workspace, Asset, and IaC Dependency Map
Goal: Connect workspace resources, assets, registries, managed identities, network restrictions, Bicep, and GitHub Actions into one deployable MLOps infrastructure map.
Tasks:
Draw a workspace-to-registry-to-deployment architecture map that shows datastore, compute, identity, network, and asset scope.
Create a command-boundary table separating Azure ML CLI, Azure CLI, GitHub CLI, and Bicep validation signals.
Write one scenario where the wrong answer changes a service name but does not fix identity, network, or version scope.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 5 - Infrastructure Failure Triage Drill
Goal: Practice diagnosing why a workspace, datastore, asset, or IaC deployment fails before model training starts.
Tasks:
Build a troubleshooting tree for datastore access failure, missing environment version, registry scope mismatch, and federated credential failure.
For each branch, name the first validation command and the expected observable result.
Convert the tree into five exam-style wrong-option patterns: broader permission, wrong scope, missing version, symptom-only monitoring, and unsupported workflow.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 6 - MLOps Infrastructure Mini Mock
Goal: Turn the full Week 1 infrastructure domain into exam-ready decision logic.
Tasks:
Write 12 mixed questions across workspace resources, assets, and IaC.
Mark each answer with the missing dependency it fixes and the verification evidence that proves it.
Update the error log with any confusion between Azure ML workspace scope, registry scope, and Azure resource-group scope.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 7 - Consolidation, Review, and Scenario Testing
Goal: Convert the week's H3 knowledge points into exam-ready decision logic.
Tasks:
Review all flashcards and scenario cards from Days 1-6.
Write 10 scenario questions that include one correct answer and three plausible distractors.
Rebuild the domain map from memory, including target objects, dependencies, failure triggers, and verification commands.
Identify three weak areas and assign next-week correction tasks.
Learning Method: Use active recall, spaced repetition, and the teaching method. No new content is required; the output is a reviewed error log, corrected flashcards, and a domain-level troubleshooting map.
Week 2 Study Plan - Implement machine learning model lifecycle and operations
Primary Objective for the Week:
Master the Implement machine learning model lifecycle and operations domain by studying its decomposed H3 knowledge points as operational workflows. By the end of the week, the learner should explain what to configure, why the configuration is required, how the exam turns it into a scenario, and which verification signal proves the answer.
Learning Methodology for the Week:
Daily duration: 3.5 to 4 hours, organized as 6 Pomodoro blocks.
Pomodoros 1-2: Knowledge Explanation reading and why-layer extraction.
Pomodoros 3-4: Diagram, command, YAML, workflow, or scenario artifact creation.
Pomodoro 5: Flashcard creation and wrong-option analysis.
Pomodoro 6: Forgetting-curve review of previous days and mixed scenario questions.
Day 1 - Orchestrate model training in Azure Machine Learning
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Orchestrate model training in Azure Machine Learning.
Tasks:
Read the Knowledge Explanation section for Orchestrate model training in Azure Machine Learning and extract Command job, MLflow run, AutoML job, Sweep job plus the failure trigger: Training output cannot be promoted when MLflow metrics, input data versions, or artifact paths are not captured.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 2 - Implement model registration and versioning
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Implement model registration and versioning.
Tasks:
Read the Knowledge Explanation section for Implement model registration and versioning and extract Registered model, MLflow artifact, Feature retrieval specification, Model version plus the failure trigger: Deployment cannot resolve the model when the registered version points to a missing artifact or incompatible feature retrieval contract.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 3 - Deploy machine learning models for production environments
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Deploy machine learning models for production environments.
Tasks:
Read the Knowledge Explanation section for Deploy machine learning models for production environments and extract Online endpoint, Online deployment, Batch endpoint, Scoring script plus the failure trigger: Requests fail when scoring schema, environment dependencies, model loading, authentication mode, or traffic routing is wrong.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 4 - Monitor and maintain machine learning models in production
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Monitor and maintain machine learning models in production.
Tasks:
Read the Knowledge Explanation section for Monitor and maintain machine learning models in production and extract Data drift monitor, Model metric, Alert rule, Retraining pipeline plus the failure trigger: Bad predictions persist when production telemetry is collected but no alert, threshold, or retraining action is bound to it.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 5 - Training-to-Deployment Lifecycle Map
Goal: Connect training jobs, MLflow tracking, model registration, endpoint deployment, monitoring, drift response, and retraining into one operational lifecycle.
Tasks:
Draw the path from training input data to job output, model artifact, registered model version, endpoint deployment, metric stream, alert, and retraining pipeline.
Create a table that separates model quality evidence, deployment health evidence, and traffic-routing evidence.
Write one scenario where the answer must choose between retraining, rollback, endpoint log inspection, or traffic adjustment.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 6 - Endpoint and Model Quality Troubleshooting
Goal: Practice distinguishing serving failure, model artifact failure, traffic routing failure, and prediction-quality degradation.
Tasks:
Build a decision tree for 401/403, environment load failure, schema mismatch, high latency, drift, and bad predictions.
Map each symptom to a validation action: endpoint show, deployment logs, invoke, metrics, model version inspection, or retraining job status.
Create five flashcards that contrast endpoint provisioning state with model performance state.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 7 - Consolidation, Review, and Scenario Testing
Goal: Convert the week's H3 knowledge points into exam-ready decision logic.
Tasks:
Review all flashcards and scenario cards from Days 1-6.
Write 10 scenario questions that include one correct answer and three plausible distractors.
Rebuild the domain map from memory, including target objects, dependencies, failure triggers, and verification commands.
Identify three weak areas and assign next-week correction tasks.
Learning Method: Use active recall, spaced repetition, and the teaching method. No new content is required; the output is a reviewed error log, corrected flashcards, and a domain-level troubleshooting map.
Week 3 Study Plan - Design and implement a GenAIOps infrastructure
Primary Objective for the Week:
Master the Design and implement a GenAIOps infrastructure domain by studying its decomposed H3 knowledge points as operational workflows. By the end of the week, the learner should explain what to configure, why the configuration is required, how the exam turns it into a scenario, and which verification signal proves the answer.
Learning Methodology for the Week:
Daily duration: 3.5 to 4 hours, organized as 6 Pomodoro blocks.
Pomodoros 1-2: Knowledge Explanation reading and why-layer extraction.
Pomodoros 3-4: Diagram, command, YAML, workflow, or scenario artifact creation.
Pomodoro 5: Flashcard creation and wrong-option analysis.
Pomodoro 6: Forgetting-curve review of previous days and mixed scenario questions.
Day 1 - Implement Foundry environments and platform configuration
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Implement Foundry environments and platform configuration.
Tasks:
Read the Knowledge Explanation section for Implement Foundry environments and platform configuration and extract Microsoft Foundry resource, Microsoft Foundry project, Managed identity, RBAC assignment plus the failure trigger: GenAI applications fail before inference when project RBAC, managed identity binding, private endpoint approval, or DNS integration is incomplete.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 2 - Deploy and manage foundation models for production workloads
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Deploy and manage foundation models for production workloads.
Tasks:
Read the Knowledge Explanation section for Deploy and manage foundation models for production workloads and extract Foundation model, Model deployment, Serverless API endpoint, Managed compute option plus the failure trigger: High-volume workloads receive throttling or unstable latency when provisioned throughput and quota are not planned.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 3 - Implement prompt versioning and management with source control
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Implement prompt versioning and management with source control.
Tasks:
Read the Knowledge Explanation section for Implement prompt versioning and management with source control and extract Prompt file, Prompt variant, Evaluation dataset, Git branch plus the failure trigger: A prompt regression cannot be diagnosed when the active prompt text, model deployment, dataset, and metric output are not tied to a commit.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 4 - Microsoft Foundry Platform Boundary Map
Goal: Clarify the boundary between Microsoft Foundry project operations and Azure OpenAI or Cognitive Services resource-plane commands.
Tasks:
Create a two-column command-boundary sheet: Microsoft Foundry platform concept versus Azure OpenAI/Cognitive Services validation command.
Map RBAC, private endpoint approval, DNS resolution, model deployment state, quota, throughput, prompt versioning, and Git release tags.
Write three distractors that misuse API keys, model size, or prompt edits to solve RBAC or network failures.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 5 - Foundation Model Capacity and Prompt Release Drill
Goal: Practice production decisions involving model/version/region support, quota, provisioned throughput, prompt evaluation, and rollback.
Tasks:
Build a release checklist that includes deployment list/show, quota inspection, metric review, prompt branch, evaluation workflow, and release tag.
Write one scenario for throttling, one for unsupported model region, and one for prompt regression after release.
For each scenario, explain why the correct answer fixes the production constraint and why the other options are partial fixes.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 6 - GenAIOps Infrastructure Mini Mock
Goal: Convert Microsoft Foundry infrastructure, foundation model deployment, and prompt source control into exam-ready scenario logic.
Tasks:
Write 12 mixed questions across RBAC, private networking, model deployment capacity, quota, and prompt versioning.
Label every wrong option as scope error, credential distraction, model-selection distraction, capacity omission, or missing release evidence.
Update the final-week review list with the three most repeated GenAIOps infrastructure mistakes.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 7 - Consolidation, Review, and Scenario Testing
Goal: Convert the week's H3 knowledge points into exam-ready decision logic.
Tasks:
Review all flashcards and scenario cards from Days 1-6.
Write 10 scenario questions that include one correct answer and three plausible distractors.
Rebuild the domain map from memory, including target objects, dependencies, failure triggers, and verification commands.
Identify three weak areas and assign next-week correction tasks.
Learning Method: Use active recall, spaced repetition, and the teaching method. No new content is required; the output is a reviewed error log, corrected flashcards, and a domain-level troubleshooting map.
Week 4 Study Plan - Implement generative AI quality assurance and observability
Primary Objective for the Week:
Master the Implement generative AI quality assurance and observability domain by studying its decomposed H3 knowledge points as operational workflows. By the end of the week, the learner should explain what to configure, why the configuration is required, how the exam turns it into a scenario, and which verification signal proves the answer.
Learning Methodology for the Week:
Daily duration: 3.5 to 4 hours, organized as 6 Pomodoro blocks.
Pomodoros 1-2: Knowledge Explanation reading and why-layer extraction.
Pomodoros 3-4: Diagram, command, YAML, workflow, or scenario artifact creation.
Pomodoro 5: Flashcard creation and wrong-option analysis.
Pomodoro 6: Forgetting-curve review of previous days and mixed scenario questions.
Day 1 - Configure evaluation and validation for generative AI applications and agents
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Configure evaluation and validation for generative AI applications and agents.
Tasks:
Read the Knowledge Explanation section for Configure evaluation and validation for generative AI applications and agents and extract Evaluation dataset, Data mapping, Groundedness metric, Safety evaluator plus the failure trigger: Scores are misleading when question, context, expected answer, and labels are mapped to the wrong evaluator fields.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 2 - Implement observability for generative AI applications and agents
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Implement observability for generative AI applications and agents.
Tasks:
Read the Knowledge Explanation section for Implement observability for generative AI applications and agents and extract Trace span, Correlation ID, Latency metric, Token metric plus the failure trigger: Production incidents cannot be reconstructed when retrieval, prompt assembly, model call, tool call, and response serialization lack a shared correlation ID.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 3 - Evaluation Dataset and Metric Gate Build
Goal: Practice turning GenAI quality requirements into dataset mapping, groundedness, relevance, safety, custom metric, and CI gate decisions.
Tasks:
Design an evaluation table with input, context, expected answer, labels, metric, threshold, and release decision.
Map each metric to the failure it detects: unsupported answer, irrelevant answer, unsafe output, tool misuse, or domain-rule violation.
Write four distractors that use manual chat review, temperature changes, or generic logs instead of repeatable evaluation gates.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 4 - Observability Trace and Cost Triage
Goal: Practice diagnosing GenAI production issues through correlated traces, latency, token usage, retrieval spans, tool spans, and alert rules.
Tasks:
Draw a trace chain from user request to retrieval, prompt assembly, model call, tool call, response, token metrics, and alert threshold.
Create a symptom-to-signal table for high latency, high token cost, hallucination, retrieval miss, tool failure, and HTTP 200 with bad answer.
Write one scenario where the correct answer requires trace correlation rather than endpoint uptime monitoring.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 5 - Quality and Observability Matrix Rehearsal
Goal: Use matrix-style verification to prove that evaluation and observability are production-ready.
Tasks:
Review every Operational Skills Matrix row for the two Week 4 topics and classify each as local lab script, GitHub workflow, Azure Monitor Logs, or Azure Monitor metric verification.
Rewrite any weak verification sentence into an observable pass/fail standard.
Create 10 quick-answer prompts that ask which signal proves the issue: dataset score, trace span, metric, log query, workflow status, or alert rule.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 6 - GenAI QA Mini Mock
Goal: Convert quality assurance and observability into exam-ready scenario selection.
Tasks:
Write 12 mixed questions across evaluation mapping, groundedness, safety, custom metrics, traces, token cost, and alerts.
For every wrong answer, explain whether it is subjective review, symptom-only monitoring, wrong metric, or missing correlation.
Update the error log with repeated confusion between quality evaluation and operational observability.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 7 - Consolidation, Review, and Scenario Testing
Goal: Convert the week's H3 knowledge points into exam-ready decision logic.
Tasks:
Review all flashcards and scenario cards from Days 1-6.
Write 10 scenario questions that include one correct answer and three plausible distractors.
Rebuild the domain map from memory, including target objects, dependencies, failure triggers, and verification commands.
Identify three weak areas and assign next-week correction tasks.
Learning Method: Use active recall, spaced repetition, and the teaching method. No new content is required; the output is a reviewed error log, corrected flashcards, and a domain-level troubleshooting map.
Week 5 Study Plan - Optimize generative AI systems and model performance
Primary Objective for the Week:
Master the Optimize generative AI systems and model performance domain by studying its decomposed H3 knowledge points as operational workflows. By the end of the week, the learner should explain what to configure, why the configuration is required, how the exam turns it into a scenario, and which verification signal proves the answer.
Learning Methodology for the Week:
Daily duration: 3.5 to 4 hours, organized as 6 Pomodoro blocks.
Pomodoros 1-2: Knowledge Explanation reading and why-layer extraction.
Pomodoros 3-4: Diagram, command, YAML, workflow, or scenario artifact creation.
Pomodoro 5: Flashcard creation and wrong-option analysis.
Pomodoro 6: Forgetting-curve review of previous days and mixed scenario questions.
Day 1 - Optimize retrieval-augmented generation performance and accuracy
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Optimize retrieval-augmented generation performance and accuracy.
Tasks:
Read the Knowledge Explanation section for Optimize retrieval-augmented generation performance and accuracy and extract Chunking policy, Vector index, Embedding model, Similarity threshold plus the failure trigger: Answers become ungrounded when thresholds exclude relevant chunks, chunking breaks context, or embedding vectors are generated with a different model than the index.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 2 - Implement advanced fine-tuning and model customization
Goal: Understand the operational object, why-layer dependency, and exam-question transformation logic for Implement advanced fine-tuning and model customization.
Tasks:
Read the Knowledge Explanation section for Implement advanced fine-tuning and model customization and extract Fine-tuning dataset, Synthetic data, Validation split, Fine-tuned model plus the failure trigger: A customized model regresses in production when synthetic examples contain label noise, duplicate patterns, unsafe outputs, or no validation baseline.
Create one scenario card that states how the exam may ask the topic, how wrong options may be designed, and why the correct answer resolves the constraint.
Rehearse at least two verification signals from the Operational Skills Matrix and label their command plane or path type.
Produce one practical artifact: command drill, YAML checklist, metric map, trace map, diagram, troubleshooting tree, or comparison table.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 3 - RAG Accuracy and Retrieval Tuning Lab
Goal: Practice deciding when to tune chunking, embeddings, thresholds, hybrid search, reranking, or evaluation before changing the model.
Tasks:
Build a RAG tuning matrix that maps symptom to retrieval metric, index change, evaluation method, and rollback evidence.
Create before/after comparison rows for chunk size, overlap, embedding model, similarity threshold, hybrid search, and semantic reranking.
Write five distractors that incorrectly fine-tune, increase tokens, or switch model size when retrieval evidence is missing.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 4 - Fine-Tuning and Customization Readiness Check
Goal: Practice validating when advanced customization is appropriate and when it is being used to solve the wrong problem.
Tasks:
Create a readiness checklist for curated data, schema validation, synthetic data inspection, validation split, evaluation result, deployment metric, and rollback path.
Compare RAG tuning, prompt engineering, and fine-tuning using objective, required evidence, risk, and production validation.
Write one scenario where fine-tuning is correct and two where retrieval or prompt evaluation should happen first.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 5 - Optimization Decision Tree
Goal: Create a final decision tree for choosing RAG tuning, evaluation, A/B testing, synthetic data review, fine-tuning, or monitoring.
Tasks:
Draw a decision tree that starts with the symptom: unsupported answer, retrieval miss, format failure, latency, token cost, or repeated domain behavior failure.
Attach one verification method to every leaf node.
Convert the tree into 10 flashcards with one correct action and one tempting but wrong action.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 6 - Final Cross-Domain Exam Simulation
Goal: Validate readiness across all five AI-300 domains using mixed scenarios and error-log correction.
Tasks:
Complete a 30-question mixed mock set that covers all 14 H3 knowledge points.
Classify every missed question by missing dependency, wrong service boundary, command mismatch, or scenario objective misread.
Run a final forgetting-curve review of Day 1, Day 3, Day 7, and Week 5 notes before building the final readiness checklist.
Learning Method: Use 2 Pomodoros for reading and deconstruction, 2 Pomodoros for artifact creation, 1 Pomodoro for flashcards and wrong-option analysis, and 1 Pomodoro for same-day active recall. Apply next-day review to the previous topic before starting new work.
Day 7 - Consolidation, Review, and Scenario Testing
Goal: Convert the week's H3 knowledge points into exam-ready decision logic.
Tasks:
Review all flashcards and scenario cards from Days 1-6.
Write 10 scenario questions that include one correct answer and three plausible distractors.
Rebuild the domain map from memory, including target objects, dependencies, failure triggers, and verification commands.
Identify three weak areas and assign next-week correction tasks.
Learning Method: Use active recall, spaced repetition, and the teaching method. No new content is required; the output is a reviewed error log, corrected flashcards, and a domain-level troubleshooting map.