Chapter 01
Foundations: uncertainty and loss
Bayes, entropy, and cross-entropy as the backbone of probability and loss.
Personal Project / ML Visualizations
Machine learning, made visual.
A visual study library for machine learning, built to make core ideas easier to understand, revisit, and explain.
Chapter Library
A visual path through the ML stack.
Chapter 02
Linear algebra for ML
Vectors, dot products, eigen directions, and SVD as the geometry underneath modern ML.
Chapter 03
Classical ML and stats
Likelihood, priors, bias-variance, and regularization as one connected mental model.
Chapter 04
Neural network basics
Neurons, activations, backpropagation, and optimization as one connected story.
Chapter 05
Deep learning: update size and gradient flow
Learning rate intuition and why repeated derivatives make training stable, brittle, or explosive.
Chapter 06
Transformers and RAG
Tokenization, attention, and retrieval as three different pieces of the same modern LLM stack.
Chapter 07
Adaptation, compression, and serving
Fine-tuning, LoRA, quantization, distillation, and the deployment tradeoffs that decide what is practical.
Chapter 08
Metrics and calibration
Thresholds, probability trustworthiness, and ranking quality in the atelier format.
Chapter 09
Recommendation depth
Latent factors, two-tower retrieval, and ranking objectives that actually match recommendation surfaces.
Chapter 10
Retrieval and systems
Retrieval versus ranking, plus the offline-versus-online tradeoffs that make systems honest.
Chapter 11
Production ML systems
Train-serve skew, proxy-vs-live evaluation, and drift in the same authored language.
Chapter 12
GBDTs and tabular ML
Useful tree splits, stagewise boosting, and why tabular ensembles still dominate so much applied ML.
Chapter 13
Data and features
Leakage, missingness, and distribution shift as the hidden reasons many models fail.
Chapter 14
Generation and alignment
Temperature, sampling behavior, and why preference tuning needs a leash instead of raw reward chasing.
Chapter 15
Reinforcement learning
MDPs, value functions, TD learning, Q-learning, and DQN in one clean ladder from loop to deep control.
Chapter 16
Alignment depth
Guidance, preference tuning, and reward hacking once you go beyond the first generative intuition.