Pair programming
22 exercises across 2 tracks and 7 stages, including 3 bonus challenges. Read unfamiliar code, discuss it, make a focused change, and test it with an interviewer.
- 1Read the code
- 2Make a change
- 3Explain your decision
What role are you preparing for?
Machine learning practice
Read an unfamiliar contract, defend a metric, run an experiment that survives review, and put a model behind an API that cannot lie about what it did.
IPython and pairing foundations
Learn to read an unfamiliar contract, preserve grid state, review classifier decisions, and finish with a first leakage-safe evaluation.
Python and pairing foundations
Learn to read an unfamiliar contract, preserve grid state, review classifier decisions, and finish with a first leakage-safe evaluation.
Conway's Game of Life with OOP
Diagnose why a complete-looking Game of Life implementation corrupts a generation while it updates the board.
Inbox Safety Cutoff
Trace a working classifier evaluator, predict how its cutoff changes errors, and defend a precision–recall decision.
Trail Arrival Model Review
Build a leakage-safe offline review, compare the current model with a candidate, and check whether aggregate improvement hides a harmful trail segment.
IIProbability, statistics, and experiments
Build trustworthy decisions from classifier metrics, controlled experiments, uncertainty, and calibrated probabilities.
Probability, statistics, and experiments
Build trustworthy decisions from classifier metrics, controlled experiments, uncertainty, and calibrated probabilities.
Alice and Bob Game Engine
Find and repair three subtle state-transition bugs in a working-looking rock-paper-scissors engine.
Repair the A/B Test Analyzer
Repair the campaign-level analysis, compare company segments and budget distributions, and return a decision that preserves conflicting business evidence.
Bayesian Rate Monitor
Complete focused Beta–Bernoulli updates and reproducible posterior comparison inside an otherwise working model.
IIIModeling and product decisions
Apply evaluation skills to recommendation ranking, causal retention questions, and a leakage-free offline validation.
Modeling and product decisions
Apply evaluation skills to recommendation ranking, causal retention questions, and a leakage-free offline validation.
Neighborhood Picks
Diagnose a neighbor-based recommender whose similarity and score calculations produce plausible but incorrect rankings.
Mentor Stipend Retention Study
Diagnose a plausible before-after analysis that wrongly attributes a shared retention trend to the mentor stipend.
Validation Leak Audit
Audit an offline model evaluator for a data leak, repair it so learned preprocessing is fit on the training split only, and restore an honest validation score.
IVPutting a model behind an API
Turn a trained model into a serving boundary that validates its input, records what it decided, and stays auditable when the decision is contested.
Putting a model behind an API
Turn a trained model into a serving boundary that validates its input, records what it decided, and stays auditable when the decision is contested.
AI engineering practice
Build a grounded application first, then practise tools, evaluation, and production safeguards before moving into advanced model internals.
IBuild the grounded application
Repair LlamaIndex ingestion and RAG retrieval tests before working with agents.
Build the grounded application
Repair LlamaIndex ingestion and RAG retrieval tests before working with agents.
Debug Report Chunking Before LlamaIndex
Repair the document-cleaning boundary that prepares structured report sections for LlamaIndex.
Incident Runbook RAG with LlamaIndex
Complete a LlamaIndex ingestion and retrieval boundary that preserves citations and applies permissions before search.
IITools, evaluation, and safe delivery
Test a LangChain RAG assistant, a CrewAI support workflow, and the evaluation gate around them.
Tools, evaluation, and safe delivery
Test a LangChain RAG assistant, a CrewAI support workflow, and the evaluation gate around them.
Repair the LangChain RAG Tests
Repair tests around LangChain Runnable boundaries, citations, abstention, and customer-scoped history.
Debug a Production Support Proposal Workflow
Debug request deduplication, tenant context, and evidence validation around a supplied CrewAI adapter.
Complete the CrewAI Tool Guard
Complete the approval and idempotency boundary around a CrewAI crew’s proposed support action.
The Tuesday Prompt Tweak
Complete the evaluation harness that scores a candidate by slice and blocks a release when any slice regresses, even while the overall score improves.
IIIOperate the production application
Protect tenant memory, survive provider failures, and monitor RAG behavior that HTTP status cannot reveal.
Operate the production application
Protect tenant memory, survive provider failures, and monitor RAG behavior that HTTP status cannot reveal.
The Cache That Must Not Leak
Complete a semantic cache that reuses an answer when a new question means the same thing, and never reuses one across a tenant, model, or prompt version boundary.
Repair the AI API Reliability Tests
Repair tests for retry classification, idempotency, fallback, and request accounting.
Repair the Production RAG Monitoring Tests
Repair tests for retrieval failures, tenant-level metrics, groundedness alerts, and expected abstention.
Bonus challengesThree additional challenges
Three additional challenges
Monty Hall: Stay or Switch?
Complete a reproducible simulation, preserve the host’s information rule, and connect the observed win rates to conditional probability.
Repair the Model Ranking Tests
Repair three incorrect pytest expectations, then prove the corrected tests detect broken ranking implementations.
Cold-Start Picks
Recommend movies to a brand-new member from a user–movie rating matrix, handling the unrated (None) entries that make the new member cold.