Bezen
Log inSign up

Practice Problems

Realistic AI engineering scenarios with reference solutions and problem-specific grading rubrics.

96 problems
ProblemCategoryDifficultyTimeBest
  • A Confusion Matrix in One Pass

    Integer indexing · bincount · Reshaping · Evaluation

    NumPy·Easy·15 minEasy15m–Solve
  • Cosine Similarity Without Loops

    Broadcasting · Vectorization · Linear algebra · Numerical stability

NumPy
·
Easy
·
20 min
Easy
20m
–
Solve
  • One-Hot Encode a Label Vector

    Fancy indexing · Broadcasting · NumPy · Encoding

    NumPy·Easy·15 minEasy15m–Solve
  • Group Sums Without a GroupBy

    bincount · Scatter-add · Safe division · Vectorization

    NumPy·Medium·25 minMedium25m–Solve
  • Intersection Over Union for Every Box Pair

    Broadcasting · Clipping · Geometry · Vectorization

    NumPy·Medium·25 minMedium25m–Solve
  • Moving Average Without a Loop

    Prefix sums · Vectorization · Off-by-one reasoning · Dtype handling

    NumPy·Medium·20 minMedium20m–Solve
  • Softmax That Does Not Overflow

    Numerical stability · Broadcasting · Axis handling · Vectorization

    NumPy·Medium·25 minMedium25m–Solve
  • Standardise Features That Have Holes In Them

    Missing data · Axis handling · Numerical guards · Feature scaling

    NumPy·Medium·25 minMedium25m–Solve
  • Top-K Retrieval Without Sorting Everything

    argpartition · Vectorization · Ranking · Complexity

    NumPy·Medium·25 minMedium25m–Solve
  • Fill the Gaps in Every Row

    Accumulate tricks · Fancy indexing · Missing data · Vectorization

    NumPy·Hard·40 minHard40m–Solve
  • Monthly Active Users

    groupby · Datetime handling · Deduplication · pandas

    Pandas·Easy·15 minEasy15m–Solve
  • Join Without Letting a Dirty Dimension Table Multiply Your Rows

    merge · Deduplication · Data quality · Join semantics

    Pandas·Medium·25 minMedium25m–Solve
  • Pad Each Store's Own Calendar, Not the Global One

    Time series · GroupBy · Reindexing · Data leakage

    Pandas·Medium·30 minMedium30m–Solve
  • Pivot Event Counts, With the Zeros Included

    pivot / unstack · Reshaping · Categorical data · GroupBy

    Pandas·Medium·25 minMedium25m–Solve
  • Pull the SKU Out of a Free-Text Note

    str.extract · Regular expressions · Data cleaning · Missing data

    Pandas·Medium·20 minMedium20m–Solve
  • Revenue Per Customer From a Messy CSV

    read_csv · Filtering · GroupBy · Data cleaning

    Pandas·Medium·25 minMedium25m–Solve
  • Rolling 7-Day Spend Per Customer

    GroupBy · Time series · Rolling windows · Feature engineering

    Pandas·Medium·30 minMedium30m–Solve
  • Sessionise an Event Stream

    GroupBy · Time series · cumsum · Feature engineering

    Pandas·Medium·30 minMedium30m–Solve
  • The Top Three Per Group, Deterministically

    sort_values · GroupBy · Ranking · Tie-breaking

    Pandas·Medium·25 minMedium25m–Solve
  • A Monthly Retention Cohort Grid

    Cohort analysis · Period arithmetic · GroupBy · Dense reindexing

    Pandas·Hard·45 minHard45m–Solve
  • Point-in-Time Feature Join

    merge_asof · Data leakage · Feature stores · Time series

    Pandas·Hard·35 minHard35m–Solve
  • Macro and Micro F1, From the Confusion Matrix Alone

    Model evaluation · Multiclass metrics · Precision/recall · F1

    Machine Learning·Easy·20 minEasy20m–Solve
  • Precision, Recall and F1 by Hand

    Model evaluation · Precision/recall · Confusion matrix · NumPy

    Machine Learning·Easy·15 minEasy15m–Solve
  • A Bootstrap Confidence Interval for a Metric

    Bootstrap · Confidence intervals · Resampling · Statistics

    Machine Learning·Medium·25 minMedium25m–Solve
  • A Pipeline That Does Not Leak

    Pipelines · Data leakage · Cross-validation · Preprocessing

    Machine Learning·Medium·30 minMedium30m–Solve
  • A Split That Does Not Leak Across Groups

    Data leakage · Cross-validation · Grouped data · Evaluation

    Machine Learning·Medium·30 minMedium30m–Solve
  • How Wrong Are Your Probabilities?

    Calibration · Model evaluation · Probability · Binning

    Machine Learning·Medium·25 minMedium25m–Solve
  • k-Nearest Neighbours, With a Tie-Break That Never Changes

    k-NN · Distance metrics · Tie-breaking · Vectorization

    Machine Learning·Medium·25 minMedium25m–Solve
  • Permutation Importance, Averaged Honestly

    Feature importance · Permutation testing · Reproducibility · Model evaluation

    Machine Learning·Medium·25 minMedium25m–Solve
  • Pick a Threshold for an Imbalanced Classifier

    Imbalanced data · Model evaluation · Precision/recall · Thresholding

    Machine Learning·Medium·25 minMedium25m–Solve
  • The Best Single-Feature Split, by Gini

    Decision trees · Gini impurity · Vectorization · Cumulative counts

    Machine Learning·Medium·30 minMedium30m–Solve
  • Would You Notice the Input Drifting?

    Drift detection · Monitoring · Quantile binning · Model maintenance

    Machine Learning·Medium·30 minMedium30m–Solve
  • k-Means, and the Cluster That Loses All Its Points

    Clustering · k-means · Convergence · Vectorization

    Machine Learning·Hard·45 minHard45m–Solve
  • ReLU, Forward and Backward

    Backpropagation · Activations · NumPy · Gradients

    Deep Learning·Easy·15 minEasy15m–Solve
  • Adam, One Step at a Time

    Optimisers · Adam · Bias correction · Training dynamics

    Deep Learning·Medium·25 minMedium25m–Solve
  • Batch Norm Behaves Differently at Inference

    Batch normalisation · Running statistics · Train/eval mode · NumPy

    Deep Learning·Medium·30 minMedium30m–Solve
  • Linear Layer Forward and Backward

    Backpropagation · Chain rule · Matrix calculus · Gradient checking

    Deep Learning·Medium·35 minMedium35m–Solve
  • Layer Norm, Backwards

    Backpropagation · Layer normalisation · Chain rule · NumPy

    Deep Learning·Hard·40 minHard40m–Solve
  • Scaled Dot-Product Attention

    Attention · Transformers · Masking · Numerical stability

    Deep Learning·Hard·40 minHard40m–Solve
  • Count the Common N-grams

    Tokenisation · N-grams · Counting · Deterministic ordering

    NLP·Easy·15 minEasy15m–Solve
  • Chunk a Document for Retrieval

    Chunking · RAG · Text processing · Sliding window

    NLP·Medium·25 minMedium25m–Solve
  • Chunk-Level Named Entity Merging

    Named entity recognition · BIO tagging · Span extraction · Sequence labelling

    NLP·Medium·25 minMedium25m–Solve
  • TF-IDF From Scratch

    TF-IDF · Sparse representations · Lexical retrieval · Vectorization

    NLP·Medium·30 minMedium30m–Solve
  • Byte-Pair Encoding Tokenizer

    Tokenization · BPE · Subword models · Vocabulary construction

    NLP·Hard·40 minHard40m–Solve
  • Chunk a Document by Size

    RAG · Chunking · Overlap · Retrieval

    LLM / GenAI·Easy·15 minEasy15m–Solve
  • Reciprocal Rank Fusion

    Hybrid search · Rank fusion · Retrieval · RAG

    LLM / GenAI·Easy·20 minEasy20m–Solve
  • A Semantic Cache That Forgets Properly

    Caching · Cosine similarity · LRU eviction · Cost control

    LLM / GenAI·Medium·30 minMedium30m–Solve
  • Fit the Conversation in the Window

    Context window · Token budgeting · Chat applications · Prompt construction

    LLM / GenAI·Medium·25 minMedium25m–Solve
  • Retrieve Five Passages, Not One Passage Five Times

    RAG · Maximal marginal relevance · Retrieval · Cosine similarity

    LLM / GenAI·Medium·30 minMedium30m–Solve
  • Structure-Aware Chunker With Overlap

    Chunking · RAG · Context construction · Retrieval quality

    LLM / GenAI·Medium·30 minMedium30m–Solve
  • Reassemble a Streamed Response

    Streaming · Server-sent events · Parsing · LLM plumbing

    AI Engineering·Easy·15 minEasy15m–Solve
  • A Rate Limiter That Allows a Burst

    Rate limiting · Token bucket · Backpressure · API design

    AI Engineering·Medium·25 minMedium25m–Solve
  • Batch Requests Without Blowing the Latency Budget

    Dynamic batching · Inference serving · Latency budgets · Throughput

    AI Engineering·Medium·30 minMedium30m–Solve
  • Retry With Exponential Backoff and Jitter

    Retries · Backoff · Thundering herd · Error handling

    AI Engineering·Medium·25 minMedium25m–Solve
  • Stop Calling the Provider That Is Down

    Circuit breaker · Resilience · State machines · Failure handling

    AI Engineering·Medium·30 minMedium30m–Solve
  • Token Cost From a Raw Usage Log

    Cost tracking · Data cleaning · CSV parsing · Normalization

    AI Engineering·Medium·25 minMedium25m–Solve
  • Token-Budgeted Context Packer

    Context management · Token budgets · RAG · Prompt construction

    AI Engineering·Medium·25 minMedium25m–Solve
  • Design a URL Shortener

    Capacity estimation · Key generation · Caching · Read-heavy systems

    ML System Design·Easy·35 minEasy35m–Design
  • Diagnose Model Drift

    Drift Analysis · Root Cause Analysis · Covariate Shift · Concept Drift

    ML System Design·Easy·25 minEasy25m–Design
  • Design a Demand Forecasting Platform

    Time Series · Hierarchical Forecasting · Backtesting · Intermittent Demand

    ML System Design·Medium·40 minMedium40m–Design
  • Design a Large-Scale Classification Platform

    Extreme Multi-Label · Hierarchical Classification · Active Learning · Human-in-the-Loop

    ML System Design·Medium·40 minMedium40m–Design
  • Design a Predictive Maintenance System

    Sensor Data · Survival Analysis · Rare Events · Edge Inference · Time Windows

    ML System Design·Medium·40 minMedium40m–Design
  • Design a Real-Time Anomaly Detection System

    Time Series · Unsupervised Learning · Streaming · Alerting · Seasonality

    ML System Design·Medium·40 minMedium40m–Design
  • Design a Recommendation System

    Candidate Generation · Ranking · Embeddings · Two-Tower Models · Cold Start

    ML System Design·Medium·45 minMedium45m–Design
  • Design an Image Search System

    Vision Embeddings · ANN Search · Multimodal · Index Sharding · CLIP

    ML System Design·Medium·40 minMedium40m–Design
  • Design an ML Training Pipeline

    Pipelines · Reproducibility · Experiment Tracking · Data Versioning · Orchestration

    ML System Design·Medium·35 minMedium35m–Design
  • Design Feature Monitoring

    Data Quality · Skew Detection · Lineage · Freshness SLAs · Root Cause

    ML System Design·Medium·30 minMedium30m–Design
  • Design Model Monitoring

    Drift Detection · Delayed Labels · Alerting · Segment Analysis · Observability

    ML System Design·Medium·35 minMedium35m–Design
  • Diagnose Training-Serving Skew

    Skew Debugging · Feature Pipelines · Data Leakage · Root Cause Analysis

    ML System Design·Medium·30 minMedium30m–Design
  • Improve a Low-Recall Model

    Imbalanced Data · Threshold Tuning · Error Analysis · Precision-Recall Tradeoff

    ML System Design·Medium·30 minMedium30m–Design
  • Debug a Production ML Model

    Incident Response · Root Cause Analysis · Debugging · Systems Thinking

    ML System Design·Hard·35 minHard35m–Design
  • Design a Model Serving Platform

    Model Serving · GPU Scheduling · Autoscaling · Canary Deploys · Multi-tenancy

    ML System Design·Hard·45 minHard45m–Design
  • Design a Personalized News Feed

    Ranking · Fan-out · Multi-Objective Optimization · Freshness · Feedback Loops

    ML System Design·Hard·45 minHard45m–Design
  • Design a Real-Time Fraud Detection System

    Streaming · Imbalanced Data · Feature Store · Low Latency · Adversarial ML

    ML System Design·Hard·50 minHard50m–Design
  • Design a Search Ranking System

    Learning to Rank · BM25 · Query Understanding · Click Models · Position Bias

    ML System Design·Hard·45 minHard45m–Design
  • Design an Ad Ranking System

    CTR Prediction · Auctions · Calibration · Delayed Feedback · Budget Pacing

    ML System Design·Hard·50 minHard50m–Design
  • Design an ML Feature Store

    Feature Store · Point-in-Time Correctness · Streaming · Data Engineering

    ML System Design·Hard·45 minHard45m–Design
  • Design Continuous ML Deployment

    CI/CD · Canary Analysis · Shadow Deploys · Rollback · Automated Promotion

    ML System Design·Hard·40 minHard40m–Design
  • Design a ChatGPT-like Application

    Streaming · Context Management · Conversation State · Safety · Cost Control

    LLM System Design·Medium·40 minMedium40m–Design
  • Design a Self-Serve LLM Fine-Tuning Platform

    GPU scheduling · Model registry · LoRA · Eval gating · Multi-tenancy

    LLM System Design·Medium·40 minMedium40m–Design
  • Design an LLM Gateway

    API Gateway · Rate Limiting · Caching · Observability · Failover

    LLM System Design·Medium·40 minMedium40m–Design
  • Design an LLM Evaluation Platform

    LLM-as-Judge · Eval Design · Regression Testing · Human Labeling · Statistics

    LLM System Design·Hard·45 minHard45m–Design
  • Design an LLM Inference Serving System

    Continuous batching · KV cache · GPU autoscaling · Quantization · Latency budgets

    LLM System Design·Hard·45 minHard45m–Design
  • Design a Document Intelligence Platform

    OCR · Structured Extraction · Confidence Calibration · Human Review · Multimodal

    RAG·Medium·40 minMedium40m–Design
  • Design a Semantic Search System

    Embeddings · Hybrid Search · Chunking · Reranking · Index Design

    RAG·Medium·35 minMedium35m–Design
  • Design a Production RAG System

    RAG · Embeddings · Vector DB · Reranking · LLM Evaluation · Access Control

    RAG·Hard·45 minHard45m–Design
  • Design an Enterprise Knowledge Assistant

    RAG · Tool Use · Personalization · Enterprise Integration · Guardrails

    RAG·Hard·45 minHard45m–Design
  • Design an AI Agent for Customer Refunds

    Tool calling · Irreversible actions · Policy enforcement · Escalation

    AI Agents·Medium·40 minMedium40m–Design
  • Design a Tool-Calling Agent Over 500 APIs

    Tool selection · Retrieval · Permissions · Failure handling · Observability

    AI Agents·Hard·45 minHard45m–Design
  • Design an AI Coding Assistant

    Code Context · Low Latency · Fill-in-the-Middle · Repo Indexing · Acceptance Metrics

    AI Agents·Hard·45 minHard45m–Design
  • Design an AI Customer Support Agent

    Agents · Tool Use · Escalation · Safety · Containment Rate

    AI Agents·Hard·45 minHard45m–Design
  • Design a Multi-Agent AI System

    Agent Orchestration · Planning · State Management · Error Recovery · Cost Control

    Agentic AI·Hard·50 minHard50m–Design
  • Design a Multi-Agent Data Analyst

    Multi-agent · Text-to-SQL · Sandboxing · Verification · Human-in-the-loop

    Agentic AI·Hard·50 minHard50m–Design
  • Design a Multi-Agent Incident Response System

    Multi-agent orchestration · Observability · Irreversible actions · Coordination

    Agentic AI·Hard·50 minHard50m–Design
  • Design a Self-Correcting Code Agent

    Coding agents · Repository context · Verification loops · Sandboxing

    Agentic AI·Hard·50 minHard50m–Design
  • Design an Autonomous Research Agent

    Agent planning · Tool use · Verification · Termination · Cost control

    Agentic AI·Hard·50 minHard50m–Design