Master Machine Learning,
from probability to agents.
Twelve connected modules that take you from your first hypothesis test to production-grade agentic AI systems — with interactive charts, worked formulas, and check-yourself quizzes in every module. No fluff, no dead links, no broken graphs.
00Start here
New to the field? Read the modules in order — each one builds on the last. Already working? Jump straight to what you need; every module is self-contained.
Statistical Foundations & Probability
Descriptive stats, distributions, Bayes, hypothesis testing, A/B tests.
Exploratory Data Analysis & Visualization
Distributions, correlation vs causation, feature triage, plotting grammar.
Supervised Learning
Regression to gradient boosting — bias/variance, metrics, model families.
Unsupervised Learning
Clustering, PCA/UMAP, anomaly detection, association rules.
Time Series & Forecasting
Stationarity, ARIMA, Prophet, deep forecasting, backtesting.
Model Tuning, Validation & Feature Engineering
CV strategies, hyperparameter search, leakage, imbalanced data, calibration.
Deep Learning & Generative Models
MLPs to diffusion & transformers; training dynamics; regularization.
Reinforcement Learning
MDPs, Q-learning, policy gradients, RLHF connection, agentic RL.
LLM Foundations: Transformers to Training
Attention, tokenization, pretraining, scaling laws, PEFT, inference serving.
LLMs II: Alignment, RAG & Multimodal
SFT → RLHF → DPO/GRPO, prompt engineering, RAG pipelines, evals, safety.
Agentic AI Systems
Tool use, MCP, planning, memory, multi-agent orchestration, agent evals.
MLOps & Production ML
Experiment tracking, CI/CD for ML, monitoring, drift, LLM serving in prod.
✦How to use this hub
Every formula is rendered with KaTeX and every chart is drawn from real data. When you hit a quiz, commit to an answer before revealing the explanation.
After each module, close the page and re-derive the key results on paper. If you can't, you've found exactly what to re-read.
Each module ends with a practice pointer. One small project per module beats ten passive read-throughs.