228 study units · flexible pace
This course begins after Python and data analytics. Build, derive, debug, retrieve after a delay, and test a changed task across classical ML, neural nets, OpenCV, RL before RLHF, LLMs, calibrated decisions, RAG, agents, generative images/video/3D, and production. Lectures are generated in batches of three. Each unit can span several sessions; baseline hours exclude extra review, remediation, and specialization.
ML math -> supervised ML -> unsupervised discovery and retrieval foundations
Mental model, generalization, just-enough math, linear regression and classification from scratch, validation discipline, capstone
Turn a competent data analyst into someone who can frame an ML problem, prove generalization, derive and code linear models from scratch, evaluate them honestly, and ship a defensible capstone — before touching trees, boosters, or neural nets.
KNN, Naive Bayes, trees, forests, gradient boosting, XGBoost/LightGBM/CatBoost, SVMs, pipelines, tuning, calibration, interpretation, packaging
Become dangerous on tabular data: derive and tune every classical supervised algorithm, build leakage-safe sklearn pipelines, calibrate and interpret models, and ship a packaged tuned booster with a written model card.
PCA, clustering, anomaly detection, recommenders, embeddings, vector search, hybrid retrieval, labeling loops
Build useful systems when labels are missing: discover structure, rank items, retrieve information, and evaluate fuzzy outputs with human review loops — before any LLM-based RAG.
Neural nets from scratch -> PyTorch -> vision and multimodal perception
Micrograd, manual backprop, MLPs, initialization, optimizers, normalization, PyTorch, debugging, tracking, scaling
Close the neural-network intuition gap: build autograd and backprop from scratch, then learn PyTorch training loops, debugging, tracking, and scaling basics.
OpenCV, image processing and geometry, CNNs, attention before ViTs, detection, segmentation, self-supervised and multimodal perception
Go beyond image-classifier demos: understand modern perception architectures, multimodal embeddings, qualitative error analysis, safety, and deployment constraints.
Classic NLP -> transformer internals -> RL foundations -> post-training, evaluation, and calibrated decisions
Tokenization, n-grams, TF-IDF, word2vec, GloVe, sequence labeling, parsing, RNNs, LSTMs, seq2seq, attention, QA, summarization
Make NLP deep before transformers: understand language tasks, classic baselines, embeddings, sequence models, attention, evaluation, and linguistic error analysis.
Self-attention, transformer blocks, BERT, GPT, T5, tokenizers, scaling laws, pretraining, decoding, HF internals, PEFT, tiny GPT
Build the transformer stack from the inside out: attention, tokenizers, GPT-style models, scaling, generation, fine-tuning, and Hugging Face parity.
Bandits, MDPs, value learning, policy gradients and PPO before RLHF; SFT, DPO, calibrated decisions, evaluation and serving
Treat LLMs like serious systems: adapt them, evaluate them, test safety and factuality, understand reasoning and post-training, and make release decisions with evidence.
Retrieval, tool use, agents, observability, security, and LLM product engineering
Diffusion, flow matching, spatial generation, world models, research experiments, and production