EdVinci LogoEdVinciSign In

ML Engineer

Career Track

Train, serve, and monitor models in production.

You want to own models end to end, from raw dataset to live endpoint.

· BY THE END YOU CAN ·
  • 01Train and evaluate models honestly, on messy real data
  • 02Turn training code into reproducible, scheduled pipelines
  • 03Serve models within a latency budget and catch drift before users do
Start this track→See all 134 skills
Compare all four tracks
7Phases
134Skills covered
6Portfolio projects
8Core tools
Some programmingStarts from

Your ML Engineer career begins

A model in a notebook is not a product. The gap between one that scores well on a test set and one that serves live traffic reliably is where machine learning engineers spend their careers — and it is mostly engineering, not mathematics.

This track covers both halves. You build the mathematical and statistical foundations properly, work through classical machine learning and then deep learning in PyTorch, and learn to evaluate a model honestly rather than optimistically.

Then it turns to production: reproducible training pipelines, experiment tracking, feature stores and the training/serving skew they exist to prevent, serving infrastructure, and the monitoring that tells you a model has quietly gone stale. It is the deepest track of the four and the one with the most operational responsibility on the other side.

Not calling somebody else’s API — here you build and own the model.

The ML Engineer job

· WHAT YOU WOULD DO ·
  • Build and train models against real, imperfect data
  • Turn training code into reproducible, scheduled pipelines
  • Serve models within a latency and cost budget
  • Monitor for drift and decide when to retrain or roll back
· WHAT POSTINGS ASK FOR ·
PythonPyTorchscikit-learnFeature engineeringMLflowFeature storesModel servingDockerKubernetesDrift detectionCI/CD for ML
· WHO HIRES FOR IT ·
  • Product companies with a model in the critical path
  • Platform teams building shared ML infrastructure
  • Research-adjacent teams putting models into production
  • Industries with heavy forecasting, ranking, or risk modelling
· WHERE IT LEADS ·
  1. 1Junior ML Engineer — Trains and evaluates models against defined problems.
  2. 2Machine Learning Engineer — Owns a model from dataset to served endpoint.
  3. 3Senior ML / MLOps Engineer — Owns the training and serving platform.
  4. 4Staff ML Engineer — Sets modelling strategy and system architecture.
· THE CURRICULUM ·

Every skill in the track, in the order you learn it.

7 phases, 134 named skills. Not chapter headings — the individual things you will be able to do by the end.

  1. PHASE 1 OF 720 SKILLS

    Mathematics and Python for machine learning

    The foundations you need, without the ones you do not.

    • NumPy arrays and broadcasting
    • Vectorisation
    • Pandas indexing and joins
    • groupby and reshaping
    • Vectors and matrices
    • Matrix multiplication
    • Dot products and norms
    • Eigenvalues and eigenvectors
    • Derivatives and partials
    • The chain rule
    • Gradient descent by hand
    • Probability and Bayes
    • Common distributions
    • Expectation and variance
    • Covariance
    • Sampling and sampling bias
    • The central limit theorem
    • Hypothesis testing
    • Confidence intervals
    • Exploratory data analysis
  2. PHASE 2 OF 729 SKILLS

    Classical machine learning

    The models that still win on most real problems.

    • Supervised vs unsupervised
    • Linear regression
    • Logistic regression
    • Decision boundaries
    • Ridge, lasso and elastic net
    • Decision trees and entropy
    • Random forests and bagging
    • Gradient boosting
    • XGBoost and LightGBM
    • Support vector machines
    • Kernels
    • k-nearest neighbours
    • k-means clustering
    • Hierarchical clustering
    • PCA
    • Dimensionality reduction
    • Train/validation/test splits
    • Data leakage
    • Cross-validation strategies
    • The bias-variance trade-off
    • Class imbalance
    • Resampling and thresholds
    • Feature engineering
    • Feature encoding
    • Feature selection
    • Hyperparameter search
    • Precision, recall and F1
    • ROC-AUC and PR-AUC
    • RMSE and MAE
  3. PHASE 3 OF 725 SKILLS

    Deep learning with PyTorch

    Neural networks from first principles, then at scale.

    • Perceptrons and MLPs
    • Activation functions
    • Vanishing gradients
    • Loss functions
    • Backpropagation
    • SGD and momentum
    • Adam and AdamW
    • Learning rate schedules
    • Warmup
    • PyTorch tensors
    • Autograd
    • Datasets and DataLoaders
    • Writing a training loop
    • Dropout
    • Batch and layer norm
    • Convolutional networks
    • Recurrent networks
    • Transformers and attention
    • Positional encoding
    • Transfer learning
    • Fine-tuning
    • Mixed precision
    • Gradient accumulation
    • Distributed data parallel
    • Seeding and reproducibility
  4. PHASE 4 OF 716 SKILLS

    Data and feature infrastructure

    Where models actually break, long before the algorithm does.

    • Data versioning
    • Lineage tracking
    • Reproducible preprocessing
    • Feature engineering at scale
    • Feature stores
    • Feature registries
    • Offline vs online features
    • Training/serving skew
    • Point-in-time correctness
    • Data validation
    • Schema enforcement
    • Labelling workflows
    • Label quality
    • Missing and outlier data
    • Augmentation
    • Synthetic data
  5. PHASE 5 OF 712 SKILLS

    Training infrastructure

    Making a result something anyone can reproduce six months later.

    • Experiment tracking
    • Run comparison
    • Hyperparameter sweeps
    • Model registries
    • Versioning and staging
    • Pipeline orchestration
    • Scheduled retraining
    • GPU scheduling
    • Spot and preemptible instances
    • Checkpointing and resuming
    • Training cost control
    • Reproducible environments
  6. PHASE 6 OF 716 SKILLS

    Deploying and serving models

    Getting a trained model behind an endpoint that meets its latency budget.

    • Model packaging
    • Dependency pinning
    • FastAPI inference services
    • TorchServe
    • Triton Inference Server
    • Ray Serve
    • Batch inference
    • Real-time inference
    • ONNX export
    • Quantisation for inference
    • Pruning and distillation
    • Autoscaling
    • Load testing
    • Latency and tail latency
    • Containerising ML workloads
    • Kubernetes for serving
  7. PHASE 7 OF 716 SKILLS

    Operating models in production

    The half of the job that starts after the model ships.

    • Metrics and dashboards
    • Alerting
    • Data drift
    • Concept drift
    • Drift detection methods
    • Quality monitoring without labels
    • Retraining triggers
    • Shadow deployments
    • Canary releases
    • Model A/B testing
    • Rollback and kill switches
    • Governance and audit trails
    • Fairness and bias auditing
    • Explainability
    • Inference cost control
    • ML incident response
· WHAT YOU WILL BUILD ·

6 projects. Every one of them a running system.

Not exercises with a known answer. Each project is something you can deploy, send someone a link to, and defend in an interview.

01Starter

An end-to-end tabular model with an honest evaluation

Take a messy real dataset through cleaning, feature engineering, cross-validated model selection, and a write-up that states what the model cannot do.

Feature engineeringCross-validationMetric selectionLeakage
02Starter

An image classifier fine-tuned from a pretrained backbone

Transfer learning on a real image set, with augmentation, a proper training loop, and error analysis on the examples it gets wrong.

PyTorchTransfer learningAugmentationError analysis
03Core

A reproducible training pipeline with tracked experiments

Versioned data, a parameterised pipeline, and tracked runs — so any result can be reproduced months later by someone who is not you.

Data versioningExperiment trackingReproducibilityPipelines
04Core

A served model with latency and drift monitoring

A trained model behind a real inference API, containerised and load-tested, with dashboards for latency, throughput, and input distribution shift.

Model servingDockerLoad testingDrift detectionMonitoring
05Core

A feature pipeline that keeps training and serving in sync

One feature definition used by both the training job and the live endpoint, with point-in-time correctness and a test that fails when skew appears.

Feature storesPoint-in-time joinsSkew detectionData validation
06Capstone

A full MLOps system with automated retraining

The complete loop: scheduled retraining, a model registry with staged promotion, canary rollout, monitoring that triggers the next cycle, and a documented rollback path.

Model registryCanary releasesRetraining triggersGovernanceRollback
· THE ML ENGINEER STACK ·
PythonPyTorchscikit-learnMLflowFeature StoresModel ServingKubernetesDrift Monitoring
· PLUS THE CORE, IN EVERY TRACK ·
Data Structures & AlgorithmsSystem DesignOperating SystemsComputer NetworksGit & GitHubDockerLinuxSQL
· ROLES THIS TRACK TARGETS ·

Machine Learning Engineer · MLOps Engineer · Applied ML Engineer

· HOW THE TRACK RUNS ·

From first lesson to signed offer.

Step 1

Learn

Work through the curriculum in order, from the fundamentals to the applied work the role does every day. Written lessons, not video you cannot search.

Step 2

Build

Ship the projects. Each one is a real system with real constraints, and each one ends up somewhere you can show it.

Step 3

Practice

Solve problems in the DSA arena, take AI mock interviews, and get feedback on the answers you actually gave.

Step 4

Get hired

Build an ATS-ready resume, see how well you match each opening, close the gaps it finds, and apply without leaving the platform.

· FREQUENTLY ASKED ·

Questions people ask before starting.

How much mathematics do I really need?

Enough to know why a model behaves the way it does: linear algebra, calculus through the chain rule, probability, and statistics. The track teaches these applied to the models that use them rather than as a separate course you finish and forget.

How is this different from the AI Engineer track?

AI Engineer builds on models somebody else trained. Here you build the model: the data, the training, the evaluation, the serving, and the monitoring. This track carries more mathematics and, in the job, more on-call responsibility.

Is this a data science track?

No. Data science leans toward analysis, experimentation, and communicating findings. This is an engineering track — the deliverable is a system that serves predictions reliably, not a report.

Do I need my own GPU?

No. Most work runs on a laptop or a free hosted notebook, and the track shows how to use rented GPU time efficiently when a project needs it.

Does it cover large language models?

Yes — transformers and attention are in the deep learning phase, and fine-tuning is covered from the model side. The applied LLM product work, retrieval and agents, is the AI Engineer track.

· NOT QUITE YOU? ·

The other three tracks.

Full Stack AI Engineer

Most popular

Ship the entire product — interface, API, and the AI layer.

You want to build and launch complete AI products on your own.

The superset track: everything an AI Engineer does, plus the product around it.

You will build
  • A responsive marketing site with a real content source
  • A REST API with authentication and PostgreSQL
  • A real-time collaborative board
TypeScriptReactNext.jsNode.jsPostgreSQLLLM APIs

Roles: Full Stack Engineer · Product Engineer (AI) · Founding Engineer

Explore this track→

AI Engineer

Build products on top of foundation models.

You want to build with large language models, not train them.

Not model training or research — that is the ML Engineer track.

You will build
  • A prompt workbench with structured outputs
  • Semantic search over a document collection
  • A production RAG assistant with citations
PythonLLM APIsRAGVector DatabasesAgents & Tool CallingEvals

Roles: AI Engineer · GenAI Engineer · LLM Application Engineer

Explore this track→

Data Engineer

Move and model data at scale.

You want to build the pipelines every other team depends on.

Not dashboards and analysis — you build the systems those run on.

You will build
  • An analytics layer over a raw dataset
  • An incremental ingestion job with tests
  • A batch ELT pipeline: orchestration, transformation, warehouse
SQLPythonApache SparkAirflowdbtApache Kafka

Roles: Data Engineer · Analytics Engineer · Data Platform Engineer

Explore this track→
· ONE PLATFORM. NO LIMITS ON WHAT YOU CAN BUILD. ·

Everything you need to grow as an engineer — software engineering, AI, real-world projects, technical challenges, hands-on practice, AI-powered guidance, interview preparation, and career development.

Don’t just learn engineering. Experience it.

Learn. Build. Solve. Practice. Prepare. Get hired.

Start Learning→