Track · ~9 months · 38 nodes
AI / ML.
Python, math foundations, models, and shipping ML in production.
Foundations
Python, math, and the data tooling everything else needs.
- Python for MLThe language of ML: syntax, environments, and idioms.
- Math essentialsThe linear algebra, calculus, and stats ML actually uses.
- Data structures for MLArrays, tensors, and shapes — thinking in dimensions.
- Jupyter & ColabNotebooks, the ML workflow, and free GPUs.
- pandas & NumPyLoading, cleaning, and reshaping data the ML way.
Classical ML
The models that solve most real problems before deep learning.
- Supervised learningRegression and classification from labeled data.
- Unsupervised learningClustering and dimensionality reduction without labels.
- Model evaluationMetrics, train/test splits, and not fooling yourself.
- Feature engineeringTurning raw data into signal a model can use.
- scikit-learnThe toolkit for classical ML, end to end.
Deep learning
Neural networks and the framework to train them.
- Neural network basicsNeurons, layers, activations, and backpropagation.
- PyTorch fundamentalsTensors, autograd, and building models in PyTorch.
- Training loopsLoss, optimizers, batches, and the training cycle.
- RegularizationFighting overfitting: dropout, weight decay, early stopping.
- Transfer learningStanding on pretrained models instead of training from zero.
LLMs & GenAI
Working with the models reshaping the field.
- LLM fundamentalsTokens, context windows, and how transformers generate text.
- Prompt engineeringGetting reliable output: structure, examples, constraints.
- RAG systemsGrounding models in your own data with retrieval.
- Fine-tuning vs promptingWhen to fine-tune a model and when a prompt is enough.
- AgentsTool use, planning, and the limits of autonomous LLMs.
MLOps
Shipping models, not just training them.
- Docker for MLReproducible environments for training and serving.
- Model servingWrapping a model in an API and serving predictions.
- Monitoring modelsTracking accuracy, latency, and inputs in production.
- MLflowTracking experiments, parameters, and model versions.
- Drift detectionNoticing when the world changes and the model decays.
Cloud serving
Running models on managed platforms at sane cost.
Production patterns
The pieces real ML systems need around the model.
Beyond the basics
Research literacy, portfolio, and the job hunt.
- Reading research papersKeeping up with a field that moves every week.
- ML system designDesigning the whole system around the model.
- Portfolio projects that rankBuilding ML projects that actually impress reviewers.
- Working with AI toolsUsing AI assistants to build ML without skipping the basics.
- Get hired (ML)Portfolio, resume, and the ML interview gauntlet.
Outcomes
When you finish this track:
- Train, evaluate, and serve a model behind an API with monitoring.
- Build a RAG system that grounds an LLM in your own data.
- Design an end-to-end ML system and defend its trade-offs in an interview.
Schedule
~266 hours total.
At 8–10 hours a week, that’s about 9 months. Each stage has its own pace.
- Foundations
- Classical ML
- Deep learning
- LLMs & GenAI
- MLOps
- Cloud serving
- Production patterns
- Beyond the basics