ML Engineering · Beginner to Expert

The full journey
from first model to
production system.

8 lessons. 123 chapters. 5 capstone projects. Master everything from Python fundamentals and data engineering to model serving, MLOps automation, cloud infrastructure, and senior-level system design.

8 lessons
123 chapters
861 learning sections
3,690+ questions
5 capstone projects
€800  ·  Lifetime access
training_run.py — epoch 0/60
Loss
2.301
Accuracy
0.000
F1 Score
0.000
AUC-ROC
0.500
Initializing training run…
Who it's for

One bootcamp, four entry points.

The lessons are unlocked from day one. Start exactly where you are.

🚀

Complete beginners

Lesson 1 assumes zero prior programming knowledge. You'll learn Python, Linux and the shell before you touch a model — and every step has exercises and 30-question assessments.

🔬

Data scientists

You can build models. Lessons 4–7 show you how to package, serve, monitor and operate them reliably — the gap between notebook and production that most data-science courses skip.

⚙️

Software engineers

You know how to code and deploy services. Lessons 1–4 fast-track you through ML fundamentals and pipelines, then Lessons 5–8 apply your existing skills to the ML-specific challenges.

🏗️

DevOps / platform engineers

You run infrastructure. Lessons 6–8 extend your Kubernetes and cloud skills directly into ML serving, MLOps automation, and senior-level ML system design — a natural progression.

The 8-lesson journey

Every stage of the ML engineering lifecycle.

Structured as a linear progression from foundations to senior architecture, with 15–16 chapters and 3 projects per lesson.

LESSON 1 · 15 CHAPTERS
Foundations of ML Engineering
OS, Python, Git, NumPy, SQL, testing, environments, logging — the complete baseline every ML engineer needs before touching a model.
Python Linux Git NumPy scikit-learn
LESSON 2 · 15 CHAPTERS
Data Engineering for ML
Storage models, ingestion patterns, ETL vs ELT, data quality, feature engineering, feature stores, data versioning, streaming vs batch — the data layer every production model depends on.
DVC Feast Pandas Parquet Kafka
LESSON 3 · 16 CHAPTERS
Model Development & Evaluation
Regression, ensembles, deep learning, cross-validation, error analysis, interpretability, transfer learning, evaluation metrics, and model documentation.
PyTorch scikit-learn SHAP LIME HuggingFace
LESSON 4 · 15 CHAPTERS
ML Pipelines, Experiment Tracking & Orchestration
scikit-learn pipelines, MLflow for experiment tracking and model registry, Airflow for orchestration, reproducibility strategies, and pipeline testing.
MLflow Airflow Hydra DVC Pytest
LESSON 5 · 15 CHAPTERS
Model Packaging, Serving & Deployment
REST and gRPC APIs, FastAPI fundamentals, Docker for ML, batch scoring, latency optimization, model compression, canary releases, and LLM serving.
FastAPI Docker ONNX TorchServe vLLM
LESSON 6 · 15 CHAPTERS
Cloud Infrastructure, IaC & Kubernetes
AWS/GCP/Azure fundamentals, IAM, networking, object storage, Terraform, Kubernetes for ML workloads, GitOps, cost optimization, and capacity planning.
AWS Terraform Kubernetes Argo CD SageMaker
LESSON 7 · 16 CHAPTERS
MLOps Automation & Production Operations
CI/CD for ML with GitHub Actions, Prometheus and Grafana monitoring, drift detection, alerting and SLOs, incident response, root-cause analysis, and ML technical debt.
GitHub Actions Prometheus Grafana Evidently PagerDuty
LESSON 8 · 16 CHAPTERS
Security, Governance & Senior ML Architecture
Threat modeling for ML, secrets management, supply-chain security, responsible AI, reliability engineering, scalability patterns, and senior-level system design with worked architecture dossiers.
OWASP Vault OPA SLOs ADRs
Curriculum

Every chapter, up front.

No marketing copy where chapter detail should be. Every chapter has 5 content sections, a practice exercise, and a 30-question assessment.

LESSON 1
Foundations of ML Engineering
15 chapters
1.1 Roles and Responsibilities
1.2 Operating Systems, Processes, and the Filesystem
1.3 Why Version Control Exists
1.4 Running Python: Interpreters, Scripts, and the REPL
1.5 Functions, Arguments, Scope, and Return Values
1.6 Why Environments Matter
1.7 Arrays and Vectorized Computation with NumPy
1.8 Relational Model: Tables, Keys, and Schemas
1.9 Why Testing Matters
1.10 What Learning from Data Means
1.11 The scikit-learn API
1.12 Sources of Non-Determinism in ML Code
1.13 Why Notebooks Don't Scale
1.14 Why Logging Matters
1.15 Documentation That People Actually Use
LESSON 2
Data Engineering for ML
15 chapters
2.1 Where ML Data Comes From
2.2 Storage Models: Row, Column, Key-Value, and Object Stores
2.3 Batch Versus Incremental Ingestion
2.4 ETL Versus ELT and Where Transformation Should Happen
2.5 Dimensions of Data Quality
2.6 Validation Concepts: Schema, Constraints, and Statistical Checks
2.7 Splitting Strategies: Random, Stratified, Grouped, and Temporal
2.8 What Features Are and Why They Matter
2.9 Training-Serving Skew: Causes and Consequences
2.10 The Problem Feature Stores Solve
2.11 Why Data Versioning Matters
2.12 Batch Versus Streaming: Latency, Complexity, and Cost
2.13 Why Distribution Is Needed: Data Volume, Memory, and Time
2.14 Personal and Sensitive Data: Classification and Risk
2.15 Taxonomy of Data Incidents
LESSON 3
Model Development & Evaluation
16 chapters
3.1 Framing Business Problems as ML Tasks
3.2 Linear and Logistic Regression in Practice
3.3 Why Ensembles Work
3.4 Classification Metrics
3.5 Holdout, k-Fold, and Stratified Cross-Validation
3.6 Parameters Versus Hyperparameters
3.7 Error Analysis Workflow
3.8 Probabilities Versus Scores
3.9 Imbalance: Effects on Training and Metrics
3.10 Neurons, Layers, and Activation Functions
3.11 Training Loops, Batching, and Data Loaders
3.12 Why Transfer Learning Works
3.13 Representing Text: Tokenization, Bag-of-Words, and TF-IDF
3.14 Global and Local Interpretability Methods
3.15 Why Offline Metrics Disagree with Production Outcomes
3.16 What Good Model Documentation Contains
LESSON 4
ML Pipelines, Experiment Tracking & Orchestration
15 chapters
4.1 From Scripts to Pipelines: Why and When
4.2 Scikit-Learn Pipelines and Custom Transformers
4.3 Why Configuration Matters in ML Systems
4.4 The Experiment-Tracking Problem
4.5 What a Model Artifact Contains
4.6 Why Registries Exist
4.7 Why Orchestration Is Needed
4.8 Airflow Architecture: Scheduler, Executor, Workers, and Metadata Database
4.9 Failure Modes in Pipelines
4.10 Incremental Versus Full Processing
4.11 Testing Strategy for ML Pipelines
4.12 Evaluation Criteria: Operability, Scale, Ecosystem, and Team Skills
4.13 Resource Requirements of Training Jobs
4.14 Levels of Reproducibility
4.15 What to Observe in a Pipeline
LESSON 5
Model Packaging, Serving & Deployment
15 chapters
5.1 What Serving Means: From Artifact to Prediction
5.2 API Design for ML Services: REST and gRPC
5.3 FastAPI Fundamentals: Routes, Models, and Dependency Injection
5.4 Failure Modes at the Serving Boundary
5.5 Why Containers: Isolation, Portability, and Reproducibility
5.6 ML-Specific Image Challenges: Size, Native Libraries, and Model Files
5.7 Registries and Image Distribution
5.8 Batch Scoring Architectures
5.9 Performance Concepts: Latency Percentiles, Throughput, and Concurrency
5.10 Why Optimize: Cost, Latency, and Hardware Constraints
5.11 What Serving Frameworks Provide
5.12 Release Risk in ML Systems
5.13 Version Coupling in ML Systems
5.14 LLM Serving Fundamentals: Tokens, Context, and Memory
5.15 Integration Patterns: Synchronous, Asynchronous, and Event-Driven
LESSON 6
Cloud Infrastructure, IaC & Kubernetes
15 chapters
6.1 Cloud Service Models and Shared Responsibility
6.2 Identity Concepts: Users, Roles, Policies, and Trust
6.3 IP, DNS, and the Network Layers Engineers Need
6.4 Object Storage Design and Lifecycle Policies
6.5 Virtual Machines and Instance Families
6.6 What Managed ML Platforms Offer (SageMaker, Vertex AI, Azure ML)
6.7 Why Infrastructure as Code
6.8 Module Design and Reuse
6.9 Why Kubernetes and When It Is Not Needed
6.10 ConfigMaps and Environment Configuration
6.11 Jobs and CronJobs for Training and Batch Inference
6.12 Why Packaging Matters
6.13 GitOps Principles and Tooling (Argo CD and Flux Concepts)
6.14 Where ML Infrastructure Money Goes
6.15 Estimating Capacity from Workload Data
LESSON 7
MLOps Automation, Monitoring & Production Ops
16 chapters
7.1 Why MLOps Exists: Lessons from Failed ML Deployments
7.2 CI Fundamentals and What Is Different for ML
7.3 GitHub Actions Concepts: Workflows, Jobs, Runners, and Events
7.4 Delivery Workflow from Registry to Production
7.5 Why Models Decay and When to Retrain
7.6 Monitoring Versus Observability
7.7 Prometheus Architecture and the Metrics Model
7.8 Types of Drift: Covariate, Label, and Concept
7.9 The Ground-Truth Problem
7.10 SLIs, SLOs, and Error Budgets for ML Services
7.11 Feedback Loops and Their Hazards
7.12 Model Lifecycle Stages and States
7.13 Incident Lifecycle and Severity Classification
7.14 Root-Cause Analysis Techniques
7.15 Systematic Debugging in Production
7.16 Sources of ML Technical Debt
LESSON 8
Security, Governance & Senior ML Architecture
16 chapters
8.1 Security Fundamentals for ML Engineers
8.2 Why Secrets Leak and the Cost of Leaks
8.3 Encryption: At Rest and In Transit
8.4 Supply-Chain Threats in ML Systems
8.5 Privacy Principles: Minimization, Purpose Limitation, and Retention
8.6 Responsible AI Principles to Practice
8.7 Governance Frameworks and Roles
8.8 Reliability Fundamentals: Availability, Durability, and Failure Domains
8.9 Scalability Patterns for Inference and Pipelines
8.10 Unit Economics of ML Systems
8.11 Requirements Discovery and Clarification
8.12 System Design Method for ML Problems
8.13 Real-Time Architecture Patterns
8.14 Evaluation Frameworks for Build-Versus-Buy
8.15 Defining Engineering Standards and Golden Paths
8.16 Understanding Stakeholders and Their Incentives
5 final capstone projects

Production-scale projects that span the full curriculum.

Each capstone integrates 3–8 lessons, requires a full A–M specification (architecture, implementation plan, rubric, stretch goals), and is independently graded to 100 points.

CAPSTONE 1
Loan Default Risk Scoring
From raw tabular data to an evaluated, reproducible credit-risk model. Covers data engineering, model selection, evaluation, and model documentation with a reproducibility audit.
L1 L2 L3
CAPSTONE 2
Support-Ticket Triage Service
Production deployment of a text classifier. Model packaging, containerized FastAPI service, cloud deployment, and a full canary-release and rollback plan.
L3 L5 L6
CAPSTONE 3
Automated MLOps Pipeline for Demand Forecasting
End-to-end feature engineering, pipeline orchestration with Airflow, GitHub Actions CI/CD gates, and an automated retraining trigger with drift-based alert rules.
L2 L4 L7
CAPSTONE 4
Operating a Fraud-Detection Service
Monitoring, drift response, and cost optimization of a live inference service. Prometheus dashboards, SLO budgets, a simulated incident postmortem, and a cost-reduction analysis.
L5 L6 L7 L8
CAPSTONE 5
Senior-Level ML Architecture and Delivery
A learner-chosen personalization or decision platform. Full architecture dossier: three-option trade-off analysis, justified design, ADRs, stakeholder review, and a governance and responsible-AI plan.
L1–L8
Salary ranges

What ML engineers earn in Europe.

Figures from PayScale, Glassdoor, and Robert Half 2025–2026. Self-study brings you the skills; the market determines what they're worth.

Junior
€52,000 – €70,000
Mid-level
€70,000 – €95,000
Senior
€95,000 – €130,000

Sources: PayScale "Machine Learning Engineer" Germany, Glassdoor ML Engineer Berlin 2025.

Junior
€50,000 – €68,000
Mid-level
€68,000 – €92,000
Senior
€92,000 – €125,000

Sources: PayScale Netherlands ML Engineer, Glassdoor Amsterdam 2025.

Junior
€46,000 – €62,000
Mid-level
€62,000 – €85,000
Senior
€85,000 – €115,000

Sources: PayScale Belgium ML Engineer, Robert Half 2026 Technology Salary Guide Belgium.

Mentor Bob · AI tutor

Stuck on a chapter? Ask Bob.

Every lesson has a built-in AI tutor scoped to the section you're reading. Bob reads the same HTML file your browser is showing, so his answers are grounded in the actual lesson content — not a generic LLM reply.

  • ✦ Explains concepts in different ways when the first one doesn't land
  • ✦ Works through practice exercises with you step by step
  • ✦ Connects the current chapter to others in the bootcamp
  • ✦ Helps debug your code against the chapter's examples
Mentor Bob — Chapter 5.3, Section A
What's the difference between a Pydantic model and a FastAPI dependency?
Good question — they solve different problems. A BaseModel defines the shape of your request or response data and handles validation automatically: if a field is missing or the wrong type, FastAPI returns a 422 before your route function even runs.
A dependency (Depends()) is about side effects and shared setup — things like loading a model from the registry once per worker, checking an API key, or opening a database connection. The dependency runs before your function and injects its return value as an argument. You'd never swap one for the other; they compose: your route uses a Pydantic model for the body and a dependency to load the ML model.
Can the dependency call the model and return a prediction directly?
You could, but the chapter recommends against it (see Section B on failure modes). If prediction logic lives in the dependency, it becomes hard to test and hard to swap the model version independently of the API contract. Better pattern: dependency loads and returns the model object; the route function handles calling it, handling timeouts, and constructing the response. That separation also makes Chapter 4.11's pipeline-testing strategies apply directly here.
Self-study · Lifetime access · No subscription

Start the training run.

890 lesson files. 3,690 assessment questions. 5 production-scale capstone projects. Unlock everything on day one and work at your own pace.

€800  ·  One-time payment  ·  Lifetime access