AI & Machine Learning Readiness

Starting diagnostic

Find your strongest entry point.

This baseline reports three skills separately. It guides your path but never locks the Topic Library. Data quality, evaluation, and responsible AI remain required for everyone.

What outcome matters most to you?

Build projects nowSkip Python basics, work through runnable ML labs, complete two projects plus a studio deliverable, and earn a free completion certificate.Build ML modelsLearn Python, math, classical ML, neural networks, and a complete tabular workflow.Build AI applicationsLearn model interfaces, retrieval, evaluation, agents, guardrails, and serving.Master deep learning systemsGo deeper on attention, GPU IO, MoE, LoRA, distributed training, and serving SLOs.Master agentic and knowledge systemsStudy bounded loops, harnesses, RAG, ontologies, GraphRAG, evaluation, and security through a production capstone.Make better AI decisionsUnderstand capabilities, evidence, risks, and responsible product choices.

Python1. For a NumPy feature matrix with shape (20, 3), what does X.mean(axis=0) return?

One mean per featureOne scalar model scoreOne mean per exampleA train/test split

Python2. Why set a random seed in a teaching experiment?

To reproduce the same randomized operation while debuggingTo prevent overfittingTo remove the need for testsTo make all models accurate

Python3. What is the safest response to incompatible array shapes?

Ignore the warning if the output looks plausibleKeep adding dimensions until code runsConvert every value to textWrite the intended input/output shapes and test the boundary

Python4. Which test best covers a standardization function?

Only that it runs without errorsOnly exact source-code matchingNormal data plus zero-variance, missing, and wrong-shape casesOnly one large random matrix

Mathematics5. What is the derivative of (w - 2)^2 with respect to w?

2(w - 2)2w squaredw - 2

Mathematics6. If events A and B are independent, which equality holds?

P(A given B) = 1P(A) = P(B)P(A and B) = P(A)P(B)P(A or B) = 0

Mathematics7. What does a dot product combine?

Rows into a dataset splitOnly vector lengthsProbabilities into a class label automaticallyPaired components into a scalar weighted sum

Mathematics8. What does a large positive covariance suggest?

The variables tend to move above and below their means togetherThe variables are independentOne variable causes the otherBoth variables have zero variance

ML reasoning9. Why keep test data separate during model selection?

To make training loss smallerBecause test rows cannot have labelsTo estimate the selected workflow without adapting choices to those examplesTo avoid creating a baseline

ML reasoning10. A classifier catches 8 of 10 actual positives. What is recall?

0.20.88.0It cannot be measured

ML reasoning11. Which is target leakage?

Scaling from training statisticsUsing information created after the prediction time as a featureComparing with a simple baselineUsing cross-validation

ML reasoning12. When should a team prefer a deterministic rule over ML?

When the rule has more lines of codeWhen a stable rule meets the need with clearer evidence and lower riskOnly when there are fewer than 1,000 rowsNever, if an ML library exists

Developer systems13. What should happen before executing a model-proposed write?

Validate the schema and authorize the exact actionRetry until it parsesIncrease the temperatureStore credentials in the prompt

Developer systems14. What makes an API retry safe after a timeout?

Idempotency plus reconciliation of external stateA larger context windowRemoving request identifiersA longer prompt

Developer systems15. Which test best protects a model interface?

Only one successful responseChecking that JSON begins with a braceContract cases for valid, malformed, denied, timeout, and duplicate behaviorA screenshot of the prompt

Agent & knowledge systems16. What should terminate an agent loop successfully?

A confident finish actionA verified goal postconditionAny non-empty tool resultThe maximum token count

Agent & knowledge systems17. Why combine graph and vector retrieval?

They recover complementary relational and semantic evidence that must still be evaluatedIt removes entity resolutionIt guarantees factual generationGraphs always replace text

Agent & knowledge systems18. What is the correct relationship between ontology and authorization?

Ontology grants tool accessGraph facts replace caller identityOntology constrains domain meaning while identity and policy independently authorize effectsThey are the same control