Skip to content
ML to LLMs

Learning path

Guided mode: work top to bottom. Each lesson moves from intuition to visualization, math, code, and engineering, then ends with a knowledge check. Reference mode: jump straight to any lesson, model, or cheat sheet.

About 63 hours of reading in full, or about 15 hours if you switch lessons to Quick read. Short on time? Use quick read, follow a learning path, or pick a 7-day interview sprint for your role. Suggested full pace: 14 weeks at 5 to 7 hours per week, including labs and projects.

Foundations

Math, information theory, vocabulary, and how to run a valid experiment.

Module 0

Mathematical and programming foundations

0/6 done

Vectors, matrices, derivatives, probability, and the Python stack, with the shape habit that prevents most bugs.

Module 1

ML vocabulary and learning paradigms

0/2 done

The precise words for samples, features, labels, parameters, losses, and metrics, and how to pick a learning paradigm.

Module 2

The end-to-end ML workflow

0/2 done

Frame, collect, explore, split, baseline, train, evaluate, deploy, monitor, without leaking the answer.

Classical machine learning

Regression, classification, neighbors, trees, clustering, PCA, and applied ML: features, imbalance, tuning, time series.

Module 3

Regression

0/2 done

Predict continuous values with linear models, regularize them, and choose the right error metric.

Module 4

Classification

0/2 done

Predict classes and calibrated probabilities with logistic regression, then turn scores into decisions with the right metric and threshold.

Module 5

Instance-based and probabilistic models

0/3 done

Nearest neighbors, Naive Bayes, and support vector machines, three very different ideas of how to classify.

Module 6

Trees and ensembles

0/2 done

Decision trees, random forests, and gradient boosting, the strongest default family for tabular data.

Module 7

Unsupervised learning

0/2 done

Clustering and anomaly detection without labels, and how to tell a real group from an artifact.

Module 8

Dimensionality reduction

0/1 done

Compress many features into a few informative directions with PCA, and use t-SNE and UMAP carefully.

Module 24

Applied ML in practice

0/5 done

The day-to-day craft of tabular ML: engineer features without leakage, handle rare positives, tune efficiently, forecast time series, and detect anomalies.

Deep learning

Text as vectors, neural networks, PyTorch, training and debugging, architectures, generative models, vision, audio, and training at scale.

Module 9

Classical NLP

0/4 done

Turn text into numbers with tokenization, bag-of-words, n-grams, TF-IDF, and word embeddings, then extract entities and discover topics.

Module 10

Neural-network foundations

0/3 done

Neurons, layers, weights, biases, activations, and tensor shapes, the building blocks of every deep model.

Module 11

Gradient descent and backpropagation

0/2 done

How models learn: follow the gradient downhill, and compute that gradient efficiently with the chain rule.

Module 12

Training and regularization

0/2 done

Make networks generalize: weight decay, dropout, early stopping, normalization, augmentation, schedules, and tuning.

Module 13

Deep-learning architectures

0/6 done

CNNs for grids, RNNs and LSTMs for sequences, autoencoders, GANs, latent diffusion, and flow matching for generation, and transfer learning.

Module 25

Computer vision and multimodal

0/6 done

Detect, segment, and encode images; learn visual features without labels; align images with text; and connect vision encoders to language models.

Module 29

Speech and audio

0/2 done

Turn sound into model input and text: spectrograms, mel filterbanks, MFCCs, CTC, transducers, attention decoders, and word error rate.

Module 27

Training at scale

0/5 done

Train models too big or too slow for one GPU: mixed precision and clipping, gradient checkpointing, collective communication, data/tensor/pipeline parallelism, and ZeRO/FSDP sharding.

Transformers and LLMs

Attention and its variants, reinforcement learning for post-training, pretraining, MoE, fine-tuning, prompting, RAG, and evaluation.

Module 14

Transformers

0/4 done

Self-attention, multi-head attention, positional encoding, and the full transformer block.

Module 26

Reinforcement learning

0/5 done

Learn from rewards instead of labels: MDPs and the Bellman equation, Q-learning and SARSA, deep Q-networks, policy gradients, and PPO, the algorithm behind RLHF.

Module 15

Large language models

0/7 done

Tokens, pretraining, decoding, adaptation, and the limits of models that predict probable text.

Module 28

Prompting and context engineering

0/3 done

Steer a model without training it: prompts as tested specifications, few-shot examples, chain-of-thought and voting, and deciding what goes into the context window.

Module 16

Retrieval-augmented generation

0/2 done

Ground LLM answers in your documents: ingest, chunk, embed, retrieve, rerank, generate, and evaluate.

Module 17

LLM evaluation

0/1 done

Measure LLM systems layer by layer: task quality, retrieval, grounding, safety, and operations.

LLM inference and agents

KV cache, batching, quantization, serving cost, tool calling, memory, MCP, and multi-agent systems.

Module 18

LLM inference serving

0/2 done

Where the time and memory go when an LLM generates text, and how servers batch, cache, and parallelize to serve many users cheaply.

Module 19

Agents and tool use

0/5 done

Let models act: tool calling with JSON schemas, the agent loop, context and memory, MCP, workflows versus agents, and the engineering that makes agents reliable.

AI engineering and system design

Production systems, security, fairness and privacy, ML system design, monitoring, A/B testing, and projects.

Module 20

Production AI engineering

0/2 done

Architecture, reliability, observability, cost, and latency for systems built around models.

Module 21

Safety and security

0/3 done

Threat-model AI systems (prompt injection, data exfiltration, tool abuse), measure and choose fairness criteria, and protect training data with differential privacy.

Module 22

ML system design and experimentation

0/3 done

Design an ML system end to end with a repeatable framework, size a recommendation funnel against a latency budget, and prove it works with a well-powered A/B test.

Module 23

Projects and career roadmap

0/2 done

A 12-week plan and five portfolio projects that turn the course into evidence of engineering skill.