AI Interview Study Paths

Welcome to the curated study paths! With over 1,801 questions in the bank, it’s impossible to review everything in a short timeframe. We have created three distinct 2-week study plans tailored to specific roles.

How to use this plan

  1. Identify your target role: Choose the path that closest matches your upcoming interview.
  2. Stick to the daily quota: Each day provides 15-20 hand-picked, high-signal questions, requiring about 2-3 hours on weekdays and 4-5 hours on weekends.
  3. Mix breadth and depth: Some days focus on wide knowledge, while weekend mock interviews demand deep, structured answers.
  4. Use diagrams & patterns: Visualizing answers is crucial for system design rounds. We’ve indicated which diagrams and patterns to study on specific days.

Adapting the plan

Path 1: Staff Machine Learning Engineer (2 weeks)

Focus Areas: Classic ML Fundamentals (Sec 3-7), System Design, Serving, MLOps (Sec 14-16), Time Series, RecSys, Coding (Sec 24-26).

Day 1: Deep Dive into ML Fundamentals

Questions: Q66–Q75, Q84, Q89, Q91, Q101, Q114 (15 questions) Diagrams: Diagram 1 (Bias-Variance Tradeoff), Diagram 4 (ROC Curve) Patterns: Pattern ML-1 (Ensemble Methods) Time: ~2.5 hours Focus tip: Make sure you can explain the bias-variance tradeoff and regularization as if explaining it to a junior engineer.

Day 2: Statistics, Probability, & A/B Testing

Questions: Q122–Q128, Q136–Q141, Q163–Q164 (15 questions) Diagrams: Diagram 7 (A/B Testing Architecture) Patterns: Pattern STAT-2 (Bayesian A/B Testing) Time: ~2 hours Focus tip: Focus heavily on A/B testing pitfalls (Simpson’s paradox, peeking, network effects) as these are highly tested for Staff roles.

Day 3: Deep Learning Core & Optimization

Questions: Q171–Q182, Q202, Q212, Q214, Q225 (16 questions) Diagrams: Diagram 12 (Backpropagation Flow) Patterns: Pattern DL-3 (Gradient Management) Time: ~2.5 hours Focus tip: Connect architectural choices (like residual connections or normalization) directly to solving optimization problems (vanishing gradients).

Day 4: Vision & NLP (Pre-LLM) Fundamentals

Questions: Q226–Q231, Q239, Q240, Q256–Q260, Q271, Q277 (15 questions) Diagrams: Diagram 18 (Transformer Encoder vs Decoder) Patterns: Pattern NLP-1 (Embeddings Evolution) Time: ~2 hours Focus tip: Understand the transition from CNNs to ViTs and Word2Vec to Contextual Embeddings.

Day 5: Classic ML System Design - Part 1

Questions: Q556–Q565, Q570, Q572, Q575, Q580, Q582 (15 questions) Diagrams: Diagram 22 (Standard ML System Architecture) Patterns: Pattern SYS-1 (Batch vs Real-time Prediction) Time: ~3 hours Focus tip: Always start with clarifying requirements and scale before jumping into model selection.

Day 6: Classic ML System Design - Part 2

Questions: Q585–Q595, Q598, Q599, Q600 (15 questions) Diagrams: Diagram 24 (Data Leakage Prevention) Patterns: Pattern SYS-3 (Cold Start Handling) Time: ~3 hours Focus tip: Be prepared to discuss failure modes: what happens when the model goes stale or input distributions shift?

Day 7: Mock Interview & Weekly Review

Questions: Q70, Q110, Q208, Q560, Q590 (5 deep-dive questions) Diagrams: Review all week 1 diagrams Patterns: Review all week 1 patterns Time: ~4 hours (simulate a real interview environment) Focus tip: Record yourself answering a system design prompt end-to-end in 45 minutes on a whiteboard or virtual pad.

Day 8: Model Serving & Inference Optimization

Questions: Q601–Q610, Q615, Q620, Q625, Q630, Q640 (15 questions) Diagrams: Diagram 28 (Model Serving Architectures) Patterns: Pattern INF-2 (Quantization & Pruning) Time: ~2.5 hours Focus tip: Contrast Triton, TF Serving, and ONNX Runtime; know when to use GPU vs CPU for inference.

Day 9: MLOps, CI/CD, and Monitoring

Questions: Q646–Q655, Q660, Q670, Q680, Q690, Q700 (15 questions) Diagrams: Diagram 32 (MLOps Lifecycle) Patterns: Pattern OPS-4 (Shadow Deployment & Canary) Time: ~2.5 hours Focus tip: Differentiate between data drift, concept drift, and model decay, with specific metrics to monitor each.

Day 10: Time Series & Forecasting

Questions: Q941–Q955 (15 questions) Diagrams: Diagram 40 (Time Series Cross-Validation) Patterns: Pattern TS-1 (Stationarity & Differencing) Time: ~2 hours Focus tip: Focus on how time-series cross-validation differs from standard k-fold, and how to handle seasonality.

Day 11: Recommender Systems Deep Dive

Questions: Q961–Q975 (15 questions) Diagrams: Diagram 42 (Two-Tower Recommendation System) Patterns: Pattern REC-2 (Candidate Generation vs Ranking) Time: ~3 hours Focus tip: The two-stage funnel (retrieval/candidate generation -> ranking) is the holy grail of RecSys interviews.

Day 12: Coding & Algorithms for ML

Questions: Q981–Q995 (15 questions) Diagrams: N/A Patterns: Pattern CODE-1 (Vectorized Operations) Time: ~3 hours Focus tip: Write out the code for standard ML algorithms (K-means, KNN, decision tree splits) from scratch in NumPy.

Day 13: Edge Cases & Staff-Level Nuances

Questions: Q130–Q135, Q251, Q502, Q612, Q622, Q635, Q958, Q978, Q998, Q1002 (15 questions) Diagrams: Diagram 45 (System Trade-offs) Patterns: Pattern ARCH-5 (Handling Data Sparsity) Time: ~2 hours Focus tip: Staff engineers are evaluated on identifying edge cases and operational reality, not just the happy path.

Day 14: Final Mock Interview & Synthesis

Questions: Q557, Q605, Q650, Q965 (4 complex system design prompts) Diagrams: Whiteboard your own architectures Patterns: Synthesize your own cheat sheet Time: ~5 hours Focus tip: Treat this as a full 4-hour loop. Don’t look at answers until you’ve fully drawn out your designs and justified trade-offs.

Emergency 3-Day Condensed Version:

Path 2: Principal AI/ML Lead (2 weeks)

Focus Areas: Strategy, Behavioral, GenAI/LLM Depth, Agents, Safety, Enterprise Architecture (Sections 1-2, 8-13, 21-23, 27-31).

Day 1: Strategy, Vision & Build vs. Buy

Questions: Q1–Q15 (15 questions) Diagrams: Diagram 50 (AI Platform ROI Matrix) Patterns: Pattern LDR-1 (Sequencing AI Capabilities) Time: ~2.5 hours Focus tip: Frame your answers in terms of business outcomes, ROI, and technical debt.

Day 2: Technical Leadership & Behavioral

Questions: Q16–Q25, Q26–Q35 (20 questions) Diagrams: N/A Patterns: Pattern LDR-3 (Managing Up & Cross-Functional) Time: ~2 hours Focus tip: Use the STAR method, but over-index on the “Trade-offs” and “Lessons Learned” for your past projects.

Day 3: LLM & Transformer Architecture Mastery

Questions: Q286–Q305 (20 questions) Diagrams: Diagram 55 (Multi-Query vs Grouped-Query Attention) Patterns: Pattern LLM-2 (KV Cache Optimization) Time: ~3 hours Focus tip: Understand exactly where the memory bottlenecks are in training vs. inference.

Day 4: Fine-Tuning, RLHF, and PEFT

Questions: Q306–Q321, Q335–Q338 (20 questions) Diagrams: Diagram 58 (RLHF vs DPO Pipeline) Patterns: Pattern LLM-4 (LoRA and Quantization) Time: ~2.5 hours Focus tip: Be ready to justify when to use prompt engineering vs RAG vs fine-tuning.

Day 5: Advanced Prompting & Structured Outputs

Questions: Q346–Q365 (20 questions) Diagrams: Diagram 60 (Function Calling Flow) Patterns: Pattern PRMP-3 (Grammar Constrained Decoding) Time: ~2 hours Focus tip: Detail how you enforce deterministic, parseable outputs from stochastic models in production.

Day 6: RAG Pipelines & Vector Databases

Questions: Q376–Q390, Q421–Q425 (20 questions) Diagrams: Diagram 65 (Advanced RAG Architecture) Patterns: Pattern RAG-2 (Retrieve-then-Rerank) Time: ~3 hours Focus tip: The naive RAG pipeline is trivial; focus on chunking strategies, hybrid search, and resolving conflicting context.

Day 7: Mock Interview - GenAI System Design

Questions: Q496, Q498, Q503 (3 extensive design prompts) Diagrams: Draw full end-to-end GenAI architectures Patterns: Pattern ARCH-8 (Multi-Model Routing) Time: ~4 hours Focus tip: Practice narrating your design choices out loud, anticipating scaling bottlenecks.

Day 8: Agentic AI & Multi-Agent Systems

Questions: Q451–Q470 (20 questions) Diagrams: Diagram 70 (ReAct Agent Loop vs Plan-and-Execute) Patterns: Pattern AGT-1 (Agent State Management) Time: ~3 hours Focus tip: Focus on memory management and preventing infinite loops in autonomous agents.

Day 9: LLM System Design Deep Dive

Questions: Q504–Q523 (20 questions) Diagrams: Diagram 75 (LLM Gateway Architecture) Patterns: Pattern SYS-6 (Streaming & Backpressure) Time: ~3 hours Focus tip: Connect agentic concepts from yesterday into scalable microservice architectures today.

Day 10: LLM Evaluation, Safety & Governance

Questions: Q831–Q840, Q866–Q875 (20 questions) Diagrams: Diagram 80 (LLM Evaluation Harness) Patterns: Pattern EVAL-1 (LLM-as-a-Judge) Time: ~2 hours Focus tip: Differentiate between offline evaluation, online metrics, and guardrail enforcement.

Day 11: Enterprise Architecture & A2A/MCP

Questions: Q1091–Q1100, Q1141–Q1150 (20 questions) Diagrams: Diagram 85 (Model Context Protocol Integration) Patterns: Pattern ENT-2 (Multi-Tenant AI Platforms) Time: ~2.5 hours Focus tip: Understand how agents communicate securely across enterprise trust boundaries (Agent-to-Agent).

Day 12: Open-Ended Architecture Probes

Questions: Q1006–Q1025 (20 questions) Diagrams: N/A Patterns: Pattern ARCH-10 (Design Pattern Synthesis) Time: ~3 hours Focus tip: These are intentionally vague. Practice your framework for narrowing down the scope before answering.

Day 13: Rapid-Fire Depth & Cloud-Native Deployment

Questions: Q1036–Q1045, Q1181–Q1190 (20 questions) Diagrams: Diagram 90 (Cloud-Native GenAI Stack) Patterns: Pattern CLD-3 (Serverless Inference) Time: ~2 hours Focus tip: Demonstrate deep knowledge of specific cloud offerings (Bedrock, Vertex, Azure OpenAI) vs OSS deployments.

Day 14: Final Mock Interview (Leadership + Architecture)

Questions: Q33, Q57, Q517, Q1145 (Behavioral + System Design) Diagrams: Synthesize your architectural viewpoints Patterns: Review strategy matrices Time: ~5 hours Focus tip: As a Principal Lead, your answers must balance technical rigor with organizational influence and cost-awareness.

Emergency 3-Day Condensed Version:

Path 3: AI Platform / Infrastructure Engineer (2 weeks)

Focus Areas: System Design, Serving, MLOps, Data Engineering, Cloud, DevOps (Sections 14-20, 26, 31).

Day 1: ML System Architecture & Serving Basics

Questions: Q556–Q565, Q601–Q605 (15 questions) Diagrams: Diagram 22 (ML System Architecture) Patterns: Pattern SYS-1 (Batch vs Real-time) Time: ~2.5 hours Focus tip: Master the transition from a Jupyter notebook to a scalable inference endpoint.

Day 2: Inference Optimization Deep Dive

Questions: Q606–Q620 (15 questions) Diagrams: Diagram 29 (TensorRT & ONNX Pipelines) Patterns: Pattern INF-3 (Continuous Batching) Time: ~3 hours Focus tip: Understand hardware bottlenecks (memory bandwidth vs compute) and optimization techniques like FlashAttention.

Day 3: MLOps Core & Model Registry

Questions: Q646–Q660 (15 questions) Diagrams: Diagram 32 (MLOps Lifecycle) Patterns: Pattern OPS-1 (Model Versioning) Time: ~2 hours Focus tip: Clearly articulate how you link a deployed model back to its exact training data and code commit.

Day 4: CI/CD for ML & Shadow Deployments

Questions: Q661–Q675 (15 questions) Diagrams: Diagram 34 (CI/CD Pipeline for Models) Patterns: Pattern OPS-4 (Shadow Deployment) Time: ~2.5 hours Focus tip: Differentiate between software CI/CD and ML CI/CD (which includes data and model validation).

Day 5: Feature Stores & Feature Engineering

Questions: Q701–Q715 (15 questions) Diagrams: Diagram 36 (Feature Store Architecture) Patterns: Pattern FEAT-2 (Online vs Offline Stores) Time: ~2.5 hours Focus tip: Explain how a feature store prevents point-in-time leakage and standardizes features across training/serving.

Day 6: Data Engineering for AI (Pipelines & Streaming)

Questions: Q726–Q740 (15 questions) Diagrams: Diagram 38 (Kafka + Spark Streaming) Patterns: Pattern DATA-3 (Lambda Architecture) Time: ~3 hours Focus tip: Be ready to design robust streaming pipelines for real-time feature computation.

Day 7: Mock Interview - Infrastructure Design

Questions: Q590, Q625, Q685 (3 massive infra questions) Diagrams: End-to-end data+ML architecture Patterns: Synthesize pipeline patterns Time: ~4 hours Focus tip: Build a complete architecture on a whiteboard, focusing on throughput, latency, and fault tolerance.

Day 8: Data Engineering at Scale (Storage & Compute)

Questions: Q741–Q755 (15 questions) Diagrams: Diagram 39 (Data Lakehouse vs Warehouse) Patterns: Pattern DATA-4 (Parquet/Iceberg Optimization) Time: ~2.5 hours Focus tip: Contrast Delta Lake, Iceberg, and Hudi. Understand columnar storage formats deeply.

Day 9: Cloud ML Platforms (AWS, GCP, Azure)

Questions: Q766–Q780 (15 questions) Diagrams: Diagram 41 (Sagemaker vs Vertex AI) Patterns: Pattern CLD-1 (Managed vs Self-Hosted) Time: ~2.5 hours Focus tip: Compare the managed offerings of major cloud providers against deploying Kubernetes (EKS/GKE) yourself.

Day 10: DevOps, Kubernetes, and GPU Orchestration

Questions: Q796–Q810 (15 questions) Diagrams: Diagram 43 (Kubeflow / K8s Architecture) Patterns: Pattern DEV-2 (GPU Time-Slicing) Time: ~3 hours Focus tip: Understand how to schedule, scale, and monitor GPU workloads efficiently in a Kubernetes cluster.

Day 11: Monitoring, Logging, and Alerting

Questions: Q676–Q685, Q811–Q815 (15 questions) Diagrams: Diagram 44 (Observability Stack) Patterns: Pattern OPS-5 (Drift Detection) Time: ~2 hours Focus tip: Focus on Prometheus/Grafana stacks and setting actionable alert thresholds for model drift.

Day 12: Coding & Algorithms for Infrastructure

Questions: Q981–Q995 (15 questions) Diagrams: N/A Patterns: Pattern CODE-2 (Concurrency & Async) Time: ~3 hours Focus tip: Write performant code focusing on multithreading, async I/O, and efficient memory usage.

Day 13: Cloud-Native Agent & LLM Deployment

Questions: Q1181–Q1195 (15 questions) Diagrams: Diagram 90 (Cloud-Native GenAI Stack) Patterns: Pattern CLD-4 (Multi-Region LLM Routing) Time: ~2.5 hours Focus tip: Apply your infrastructure knowledge to the specific constraints of large language models (vLLM, TGI).

Day 14: Final Infra Mock Interview

Questions: Q750, Q795, Q825, Q1199 (4 infra design prompts) Diagrams: Whiteboard full platform designs Patterns: Review all scaling patterns Time: ~4 hours Focus tip: Design a platform that serves hundreds of models and processes terabytes of data daily, while maintaining strict SLAs.

Emergency 3-Day Condensed Version: