Clear explanations. Practical learning. A path that makes sense.
Certification prep for people who want to understand.
Find your certification. Follow your curiosity. Make it click.
From your first baseline to production-ready ML.
A guided journey through the complete ML lifecycle.
Read the vignette. Value the investment. Explain the decision.
Explore the CFA path ↗BigQuery ML, AutoML, or a foundation model? Learn to match the tool to the problem before you write a line of code.
Prevent leakage, build useful features, and make every experiment reproducible.
Reason about training strategies, tuning, and compute without reaching for the biggest machine.
Batch or online inference, rollout strategies, and the practical tradeoffs of model serving.
Connect validation, training, evaluation, and release into a repeatable ML workflow.
Understand drift, evaluate generative outputs, and design a safer feedback loop.
Use a confusion matrix, precision, recall, and decision thresholds to evaluate a classifier against real business costs.
Distinguish serving skew, input drift, and concept drift, then investigate alerts with an evidence-led response.
Reconstruct historical features accurately, separate event time from availability time, and design evaluations that resemble production.
Understand retrieval-augmented generation, trace failure modes, and evaluate both evidence retrieval and generated answers.
Design a serving contract, stage a canary release, and make rollback decisions using reliable evidence.
Profile training bottlenecks, design defensible tuning trials, and distinguish distributed training from parallel experiments.
Design a repeatable training pipeline with data validation, candidate comparison, promotion gates, and traceable artifacts.
Compare model outcomes across relevant slices, interpret fairness metrics, and use feature attribution as a debugging tool.
Compare warehouse models, managed model development, specialized APIs, and foundation models using concrete requirements.
Map identities, data access, notebook risks, and layered defenses for predictive and generative AI systems.
Diagnose whether an AI task needs clearer instructions, current evidence, or a change in learned behavior.
Define a forecast horizon, build time-aware baselines, and distinguish model diagnostics from future accuracy.
Choose an appropriate processing approach and reason about event time, late data, and consistent features.
Build a representative evaluation set, define rubrics, calibrate model-based judges, and investigate regressions.
Connect class imbalance, ranking, probability calibration, and threshold choice to the capacity of a real decision process.
Connect experiments, versioned models, deployment records, and rollback evidence across the ML lifecycle.
Measure latency, throughput, and cost together before changing hardware, batching, or model size.
Trace feature values from a BigQuery source to training and online prediction, with checks for freshness and consistency.
Use learning curves, data checks, and targeted experiments to investigate poor model performance.
Define labels, account for delayed and selectively observed outcomes, and build representative evaluation sets.
Understand embedding task types, approximate search, and the difference between index recall and useful evidence.
Design an online comparison with stable assignment, meaningful metrics, safety guardrails, and a predeclared analysis.
Compare data and model parallelism, diagnose scaling limits, and plan recovery for distributed training.
Choose a serving container, implement prediction and health contracts, and test model artifacts with their preprocessing.
Calculate MAE and RMSE, inspect residuals, and choose evaluation measures that reflect the cost of error.
Use clustering to explore groups while checking feature scale, stability, and real-world usefulness.
Handle missing and categorical values, prevent transformation leakage, and test parity between training and serving.
Read sample counts, paired comparisons, and slice-level metrics before deciding whether a model is better.
Choose an anomaly-detection approach, set review policies, and evaluate alerts without confusing unusual with harmful.
Compare ML-focused pipelines and broader schedulers while designing explicit dependencies, retries, and release gates.
Map sensitive data through training, evaluation, serving, and logs, then choose controls that fit the intended use.
Compare simple baselines, linear models, tree ensembles, and managed approaches against the same decision criteria.
Choose an image task, design leakage-resistant splits, and evaluate performance across real capture conditions.
Distinguish local and global explanations, interpret baselines, and investigate attribution changes responsibly.
Establish a frozen baseline, unfreeze deliberately, and keep the release reproducible.
Define the recommendation decision, diagnose the stage that failed, and evaluate the complete experience.
Write semantic contracts, separate blocking failures from signals, and make recovery explicit.
Preserve the evaluation population, distinguish weights from duplication, and report the tradeoff.
Start with the deployment constraint, choose representative calibration inputs, and release against explicit gates.
Choose the unit of evaluation, calculate reciprocal rank, and investigate disagreements.
State the invariance assumption, split originals before augmentation, and test the promised robustness.
Name every meaningful input, separate deterministic work from effects, and prove repeatability and invalidation.
Extract relevant facts, manage units and assumptions, and review mistakes with a repeatable item-set workflow.
Distinguish equity and firm cash flows, work a dividend valuation, and test how assumptions drive terminal value.
Discount cash flows with spot rates, derive a forward rate, and distinguish a rate scenario from a yield prediction.
Interpret conditional coefficients, compute a prediction, and separate statistical significance from economic usefulness.
Map the six official domains to guides, practice prompts, and areas to explore further.
Combine guided reading, hands-on exercises, and recall checkpoints in an adaptable schedule.
Review classification, regression, forecasting, and retrieval metrics with worked calculations.
Reason through six original scenarios and reveal the explanations when you are ready.
Connect data preparation, model selection, deployment, and monitoring in one fictional case.
Design authorized retrieval, evaluate evidence and answers, and apply release gates.
Distinguish a feature-pipeline defect from an unexplained change in the population.
Run a small Python exercise with separate validation and test data. No cloud account required.
Use a small Python dataset to uncover late-arriving and future-event leakage.
Inspect a public dataset, train on earlier data, and evaluate on later data.
Try another topic or search term, or save a guide to add it to your reading list.