The ML Engineer builds, trains, and ships models as production software. roadmap.sh splits this role’s body of knowledge across two roadmaps, and this track covers both — they’re genuinely different disciplines under one role:
- ML / modeling (
machine-learning) — the models themselves: the algorithm zoo, deep learning, evaluation. Sections 01–09. - MLOps / operations (
mlops) — running models in production: deployment, monitoring, pipelines, retraining. Essentially “DevOps for ML.” Section 10.
Contrast the siblings: the Data Scientist uses models to find answers (and may touch neither production nor operations); the AI Engineer builds on pre-trained models without training their own. The ML Engineer owns the model and ships it. Part of the AI/ML course.
Reading order
Modeling half:
- 01-foundations-of-ml/ — [Beginner] — what ML is, the types of learning (supervised/unsupervised/RL), and the Python/scikit-learn/PyTorch toolkit
- 02-working-with-data/ — [Intermediate] — cleaning, feature engineering & scaling, train/val/test splits (data is where most ML work actually happens)
- 03-classical-ml/ — [Intermediate] — the algorithm zoo: regression & regularization, trees & ensembles, SVM/KNN/naive Bayes, clustering & PCA
- 04-model-evaluation/ — [Intermediate] — metrics, overfitting/regularization, cross-validation, model selection, hyperparameter tuning
- 05-deep-learning/ — [Intermediate → Advanced] — neural network fundamentals (perceptron/MLP, backprop, activations, losses) and a real PyTorch training loop
- 06-computer-vision/ — [Intermediate] — CNNs, image data/augmentation, transfer learning
- 07-sequence-models-and-nlp/ — [Advanced] — RNN/LSTM/GRU, transformers & attention, NLP & embeddings
- 08-other-architectures/ — [Advanced] — autoencoders & GANs, recommendation systems, reinforcement learning
- 09-building-and-fine-tuning/ — [Advanced] — designing an architecture, training from scratch vs fine-tuning, and explainability (SHAP/LIME)
Operations half:
- 10-mlops/ — [Advanced] — the ML lifecycle in production: experiment tracking, data pipelines, serving, drift monitoring, edge AI — cross-linking the DevOps domain for the shared infrastructure
Where the content comes from
The original flat ml-engineering/computer-vision/building-your-own-models content was re-homed into this reading order (nothing dropped), and the roadmap.sh gaps filled in this pass. It’s mostly [reference] — solid conceptual coverage with real scikit-learn/PyTorch code, but the reps are doing: train the models, run the experiments, ship one. Where MLOps meets infrastructure, it cross-links DevOps rather than re-explaining Docker/k8s/CI/CD/Terraform/Prometheus.
Related
- Mathematics — the linear algebra/calculus/probability these models are built on
- DevOps — the operational tooling the MLOps section builds on
- AI/ML course map — the three paths
Software engineering
Infrastructure & security
Computer science
Hardware & engineering