The roadmap

From dot products to autonomous agents.

Seven branches, ordered by dependency. Each concept is one sentence you can expand — tick it off once you could explain it to someone else. Progress stays in your browser.

7
branches
593
concepts
555
articles live
90.4h
total reading

● Branch 01

Mathematics

The five languages every model and every analysis is written in.

0/26 known

2.5 h total

requires: high-school algebraoutcome: Read any ML paper's notation without stopping, and judge any statistical claim.
  1. 1.1Linear Algebra

  2. 1.2Calculus & Optimization

  3. 1.3Probability

  4. 1.4Statistics

  5. 1.5Information Theory

● Branch 02

ML Algorithms

Classical models — small, fast, and still everywhere.

0/68 known

6.9 h total

requires: Mathematicsoutcome: Choose, train, and honestly evaluate a model for tabular data.
  1. 2.1Supervised · Regression

  2. 2.2Supervised · Classification

  3. 2.3Unsupervised · Clustering

  4. 2.4Unsupervised · Dimensionality Reduction

  5. 2.5Unsupervised · Association Rule Learning

  6. 2.6Unsupervised · Anomaly Detection

  7. 2.7Evaluation & Generalization

● Branch 03

Deep Learning

Differentiable programs trained end to end.

0/376 known

67.4 h total

requires: ML Algorithmsoutcome: Build, train, and debug a neural network from scratch.
  1. 3.1Deep Learning Fundamentals

  2. 3.2Activation Functions

  3. 3.3Neural Network Algorithms

  4. 3.4Loss Functions

  5. 3.5Optimization Algorithms

  6. 3.6Regularization & Normalization

  7. 3.7Convolutional Neural Networks

  8. 3.8Recurrent Neural Networks and Sequence Models

  9. 3.9Attention and Transformer Architectures

  10. 3.10Generative Deep Learning Algorithms

  11. 3.11Large Language Models (LLMs)

  12. 3.12Deep Learning for NLP

  13. 3.13Deep Learning for Computer Vision

  14. 3.14Speech and Audio Deep Learning

  15. 3.15Graph Neural Networks (GNNs)

  16. 3.16Architectures

● Branch 04

Reinforcement Learning

Learning to act from reward alone, by trial and error.

0/16 known

1.4 h total

requires: Deep Learningoutcome: Frame a problem as an MDP and train an agent with value- or policy-based methods.
  1. 4.1Foundations

  2. 4.2Tabular Methods

  3. 4.3Deep RL

  4. 4.4Beyond the Basics

● Branch 05

Generative AI

Models that produce text, images, and more.

0/85 known

11.0 h total

requires: Deep Learningoutcome: Prompt, ground, customise, evaluate, secure and serve generative models in real applications.
  1. 5.1Foundations of Generative AI

  2. 5.2Prompt Engineering

  3. 5.3Decoding and Generation Control

  4. 5.4Retrieval-Augmented Generation (RAG)

  5. 5.5Multimodal Generative AI

  6. 5.6Customization

  7. 5.7Evaluation of Generative Models

  8. 5.8Safety, Alignment and Responsible AI

  9. 5.9Serving Generative Models

  10. 5.10Applications

● Branch 06

Agentic AI

Models that act — call tools, plan, and loop.

0/13 known

0.8 h total

requires: Generative AIoutcome: Design, build, and evaluate a reliable tool-using agent.
  1. 6.1LLM Agents

  2. 6.2Memory & Context

  3. 6.3Planning & Orchestration

  4. 6.4Reliability

● Branch 07

MLOps & Systems

Getting models out of notebooks and into production.

0/9 known

0.6 h total

requires: ML Algorithmsoutcome: Ship, serve, and monitor a model you can reproduce.
  1. 7.1Data & Experiments

  2. 7.2Deployment & Inference

  3. 7.3Monitoring & Scale