This guide provides a structured study plan for ML, AI Engineer, and Data Engineer interview preparation, with difficulty levels and time estimates.
| Level | Label | Description |
|---|---|---|
| 🟢 Beginner | Entry-level | Conceptual understanding; expected from all candidates |
| 🟡 Intermediate | Mid-level | Applied knowledge; expected for 2–5 YOE roles |
| 🔴 Advanced | Senior-level | Deep technical; expected for senior/staff roles |
Recommended for: All ML/AI/DS roles Estimated prep time: 2–3 weeks
| Topic | Difficulty | Key Questions to Master |
|---|---|---|
| Supervised vs Unsupervised Learning | 🟢 | Differences, use cases, examples |
| Bias-Variance Tradeoff | 🟢 | What each means, how to fix each |
| Overfitting & Regularization (L1/L2) | 🟡 | When to use Ridge vs Lasso |
| Gradient Descent (GD, SGD, Mini-batch) | 🟡 | Differences, convergence, learning rate |
| Linear & Logistic Regression | 🟢 | Assumptions, cost functions, interpretation |
| Decision Trees & Random Forests | 🟡 | Gini vs Entropy, bagging, feature importance |
| SVM | 🟡 | Kernel trick, margin, C parameter |
| Evaluation Metrics (Precision, Recall, F1, AUC-ROC) | 🟢 | When to use each, business context |
| K-Nearest Neighbors | 🟢 | Distance metrics, curse of dimensionality |
| Naive Bayes | 🟡 | Assumptions, Bayes theorem, applications |
| Boosting (XGBoost, AdaBoost) | 🟡 | Bagging vs Boosting, gradient boosting |
| Clustering (K-Means, DBSCAN) | 🟡 | Choosing k, inertia, elbow method |
| Dimensionality Reduction (PCA, t-SNE, UMAP) | 🟡 | Variance explained, visualization |
| Recommender Systems | 🔴 | Collaborative filtering, matrix factorization |
| Time Series (ARIMA, Prophet) | 🔴 | Stationarity, seasonality, forecasting |
Prerequisites: Python basics, linear algebra fundamentals
Recommended for: ML Engineer, Data Engineer, Analytics Engineer, AI Engineer Estimated prep time: 1-2 weeks in parallel with the main track
| Topic | Difficulty | Key Questions to Master | Guide |
|---|---|---|---|
| Python fundamentals for interviews | 🟢 Beginner | Lists, dicts, sets, strings, edge cases | Python Coding Challenges |
| Sliding window / two pointers | 🟡 Intermediate | Substrings, contiguous ranges, rolling windows | Python Coding Challenges |
| Trees, graphs, BFS, DFS | 🟡 Intermediate | Dependencies, traversal, cycle detection | Python Coding Challenges |
| SQL aggregations and joins | 🟢 Beginner | Grouping, nulls, output grain | SQL Coding Challenges |
| SQL windows and ranking | 🟡 Intermediate | Running totals, top-N per group, lag/lead | SQL Coding Challenges |
| SQL retention and funnel patterns | 🔴 Advanced | Cohorts, sessionization, staged CTE logic | SQL Coding Challenges |
Use this track alongside:
Recommended for: ML Engineer, DL Engineer, AI Engineer Estimated prep time: 2 weeks
| Topic | Difficulty | Key Questions to Master |
|---|---|---|
| Neural Network Architecture | 🟢 | Layers, activations, forward/backprop |
| Activation Functions (ReLU, Sigmoid, Softmax) | 🟢 | When to use each, vanishing gradient |
| Batch Normalization | 🟡 | Why it works, training vs inference |
| CNNs | 🟡 | Convolution, pooling, receptive field |
| RNNs & LSTMs | 🟡 | Sequence modeling, gating mechanism |
| Attention Mechanism & Transformers | 🔴 | Self-attention, multi-head, positional encoding |
| Transfer Learning & Fine-tuning | 🟡 | When to fine-tune vs train from scratch |
| Regularization (Dropout, Weight Decay) | 🟡 | Preventing overfitting in deep nets |
| Optimization (Adam, AdaGrad, RMSProp) | 🟡 | Adaptive learning rate methods |
Prerequisites: Classic ML Track, calculus, linear algebra
Recommended for: AI Engineer, GenAI Engineer, LLM Engineer Estimated prep time: 3–4 weeks
| Topic | Difficulty | Key Questions to Master | Guide |
|---|---|---|---|
| RAG Architecture | 🟡 | Indexing, retrieval, generation pipeline | RAG Guide |
| RAG Evaluation (RAGAS) | 🔴 | Context precision/recall, faithfulness, answer relevance | RAG Guide |
| Vector Databases | 🟡 | ANN algorithms, HNSW, choosing a vector DB | Vector DBs Guide |
| Embedding Models | 🟡 | Dense vs sparse, embedding drift | Vector DBs Guide |
| LLMOps | 🔴 | Tracing, evals, cost tracking, deployment | LLMOps Guide |
| Agentic AI | 🔴 | ReAct, Plan-Execute, multi-agent systems | Agentic AI Guide |
| n8n / AI Workflow Automation | 🟡 | Webhooks, approval flows, AI-enabled business automation | n8n Guide |
| MCP (Model Context Protocol) | 🔴 | Tool servers, client-server architecture | MCP Guide |
| LangChain / LangGraph | 🟡 | LCEL, chains, agents, memory | LangChain Guide |
| Anthropic / Claude API | 🟡 | Tool use, caching, extended thinking | Anthropic Guide |
| Fine-tuning vs RAG | 🔴 | When to use each, tradeoffs | 2026 Questions |
Prerequisites: Python, REST APIs, basic LLM familiarity
Recommended for: MLOps Engineer, Senior ML Engineer Estimated prep time: 2–3 weeks
| Topic | Difficulty | Key Questions to Master | Guide |
|---|---|---|---|
| MLflow | 🟡 | Experiment tracking, model registry | MLflow Guide |
| Feature Stores | 🔴 | Online vs offline, point-in-time correctness | Feature Stores Guide |
| Model Serving | 🔴 | Latency vs throughput, batching, scaling | Model Serving Guide |
| Model Explainability (SHAP, LIME) | 🟡 | Global vs local explanations | Explainability Guide |
| Data Quality & Validation | 🟡 | Data contracts, drift detection | Data Quality Guide |
| Model Monitoring | 🔴 | Data drift, concept drift, alerting | LLMOps Guide |
Prerequisites: Classic ML Track, Python, Docker basics
Recommended for: Data Engineer, Analytics Engineer, Platform Engineer Estimated prep time: 3–4 weeks
| Topic | Difficulty | Key Questions to Master | Guide |
|---|---|---|---|
| Data Modeling | 🔴 | Grain, keys, SCDs, dimensional vs Data Vault, modeling for BI and ML | Data Modeling Guide |
| Data Architecture | 🔴 | Batch vs streaming, lakehouse, medallion, Lambda/Kappa, governance | Data Architecture Guide |
| Apache Spark | 🟡 | RDDs vs DataFrames, partitioning, joins | Spark Guide |
| Apache Kafka | 🟡 | Topics, partitions, consumer groups, exactly-once | Kafka Guide |
| Apache Airflow | 🟡 | DAGs, operators, XComs, sensors | Airflow Guide |
| dbt | 🟡 | Models, tests, macros, incremental models | dbt Guide |
| Apache Iceberg | 🔴 | Time travel, schema evolution, merge-on-read | Iceberg Guide |
| Delta Lake | 🟡 | ACID transactions, Z-ordering, CDC | Delta Lake Guide |
| DuckDB | 🟢 | Columnar analytics, use cases vs Spark | DuckDB Guide |
Prerequisites: SQL proficiency, Python, basic cloud knowledge
Recommended for: MLOps Engineer, Platform Engineer, DevOps Engineer Estimated prep time: 2–3 weeks
| Topic | Difficulty | Key Questions to Master | Guide |
|---|---|---|---|
| Docker | 🟢 | Images, containers, networking, Dockerfile | Docker Guide |
| Kubernetes | 🟡 | Pods, deployments, services, HPA | K8s Guide |
| Helm | 🟡 | Charts, values, templating | Helm Guide |
| Terraform | 🟡 | State, modules, plan/apply | Terraform Guide |
| GitHub Actions | 🟢 | CI/CD workflows, secrets, matrix builds | GitHub Actions Guide |
Prerequisites: Linux basics, cloud platform familiarity
- Classic ML Foundations (Track 1) → 2 weeks
- Deep Learning (Track 2) → 2 weeks
- MLOps & Production (Track 4) → 1 week
- DevOps basics (Track 6: Docker, K8s) → 1 week
- Classic ML Foundations (Track 1) → 1 week (skim)
- Deep Learning — Transformers/Attention (Track 2) → 1 week
- GenAI Engineering (Track 3) → 3 weeks
- LLMOps (overlap with Track 4) → 1 week
- SQL & Python proficiency (prerequisite)
- Data Engineering tools (Track 5) → 3 weeks
- DevOps basics (Track 6) → 1 week
- Classic ML overview (Track 1) → 1 week (skim)
- Classic ML (Track 1) → 1 week
- MLOps & Production (Track 4) → 2 weeks
- Data Engineering (Track 5) → 2 weeks
- DevOps (Track 6) → 2 weeks
Essential for all roles:
- Data structures: Lists, stacks, queues, strings, hash maps, vectors, matrices, classes/objects, trees, graphs
- Algorithms: Recursion, searching, sorting, optimization, dynamic programming
- Complexity: P vs. NP, big-O notation, approximate algorithms
- Computer architecture: Memory, cache, bandwidth, threads/processes, deadlocks
- Basic probability: Conditional probability, Bayes rule, likelihood, independence
- Probabilistic models: Bayes Nets, Markov Decision Processes, Hidden Markov Models
- Statistical measures: Mean, median, mode, variance, population vs. sample statistics
- Proximity and error metrics: Cosine similarity, MSE, Manhattan/Euclidean distance, log-loss
- Distributions: Uniform, normal, binomial, Poisson
- Analysis methods: ANOVA, hypothesis testing, factor analysis
- Library calls, REST APIs, data collection endpoints, database queries
- User interface: Capturing inputs, displaying results and visualizations
- Scalability: Map-reduce, distributed processing
- Deployment: Cloud hosting, containers, microservices
- Think out loud — interviewers want to follow your reasoning, not just the answer
- Clarify before you code — ask about constraints, edge cases, scale requirements
- Start simple — give a naive/brute-force answer first, then optimize
- Know your tradeoffs — every algorithm has pros and cons; be ready to discuss them
- Bridge theory to practice — relate concepts to real systems (e.g., "In production, I would...")
- Admit uncertainty honestly — "I'd need to verify this, but I believe..." is better than guessing confidently
See 2026 Interview Roadmap for the latest focus areas. See Resources and References for books and external learning materials.