Build systems that learn
Move from Python and data fundamentals to trained models, modern AI applications, evaluation, and production deployment.
The learning path
Do not treat the resource links as a reading list. Learn enough to produce the outcome, then move forward.
Milestone 1
Learn to express problems in code and build intuition for the algebra, probability, and calculus behind machine learning.
CheckpointAnalyze a dataset with vectorized Python and explain the math behind your transformations.
Milestone 2
Turn raw data into reliable inputs through cleaning, feature engineering, visualization, and reproducible notebooks.
CheckpointCreate a reproducible exploratory analysis with a clear data-quality report.
Milestone 3
Frame prediction problems, train baselines, choose useful metrics, and avoid leakage or misleading validation.
CheckpointTrain and compare several models using a defensible evaluation protocol.
Milestone 4
Understand tensors, optimization, neural architectures, and the training loop before reaching for large models.
CheckpointImplement, train, debug, and document a neural network from a clean training loop.
Milestone 5
Build with pretrained models, embeddings, retrieval, structured outputs, tool use, and responsible prompt design.
CheckpointBuild a grounded AI workflow that cites sources and handles failure cases.
Milestone 6
Treat models as living systems with versioned data, evaluation suites, monitoring, deployment, and rollback plans.
CheckpointDeploy a model-backed service with automated evaluation and operational metrics.
Capstone project
Ship an AI research assistant with retrieval, evaluation, observability, and a documented model card.