The AI / ML roadmap, five lines at a time.

Every concept from what a dataset is through to reproducing research papers, explained in five lines and graded as you go. Each module is numbered in the order it is meant to be read; every one of the 356 concepts has its five lines free.

356Concepts

17Modules

21Graded tests

430Questions

Start studying →

Jump to a module

Machine Learning

55 concepts · 55 questions Study module 01 →

How machines learn from data, the main algorithm families, and how to measure whether any of it worked.

28 basics25 how it works2 the edge

Deep Learning

25 concepts · 25 questions Study module 02 →

Neural networks from a single neuron up to the training tricks that make deep models actually converge.

11 basics12 how it works2 the edge

Computer Vision

22 concepts · 22 questions Study module 03 →

Getting machines to read images: convolution, CNN architectures, detection, segmentation and vision transformers.

16 basics5 how it works1 the edge

NLP

20 concepts · 20 questions Study module 04 →

Turning language into numbers, and the recurrent architectures that handled sequences before transformers arrived.

11 basics6 how it works3 the edge

Transformers

19 concepts · 24 questions Study module 05 →

The attention mechanism piece by piece, and how the encoder, decoder and decoder-only stacks are assembled from it.

10 basics7 how it works2 the edge

LLMs

22 concepts · 22 questions Study module 06 →

What a large language model actually does at each step, and the sampling knobs that shape what comes out.

11 basics8 how it works3 the edge

Prompt Engineering

17 concepts · 17 questions Study module 07 →

Structuring what you send a model so the output is reliable, parseable and hard to hijack.

10 basics6 how it works1 the edge

Embeddings & Search

17 concepts · 17 questions Study module 08 →

Representing meaning as vectors, and the index structures that find the nearest ones fast enough to matter.

10 basics6 how it works1 the edge

RAG

19 concepts · 21 questions Study module 09 →

Grounding a model in your own documents: the full pipeline from raw files through to a cited answer.

8 basics6 how it works5 the edge

Fine-Tuning

15 concepts · 15 questions Study module 10 →

Changing the weights rather than the prompt, and knowing when that is actually the right lever.

5 basics8 how it works2 the edge

AI Agents

17 concepts · 17 questions Study module 11 →

Models that take actions in a loop: tools, planning, memory, and knowing when to hand control back to a person.

7 basics8 how it works2 the edge

Generative AI

18 concepts · 19 questions Study module 12 →

Models that create rather than classify: autoencoders, GANs, diffusion, and the multimodal systems built on them.

10 basics6 how it works2 the edge

AI Evaluation

18 concepts · 18 questions Study module 13 →

Deciding whether the thing you built actually works, especially when there is no single correct answer.

7 basics10 how it works1 the edge

MLOps / LLMOps

22 concepts · 23 questions Study module 14 →

Everything between a working notebook and a service other people depend on: shipping, tracking, watching and paying for it.

9 basics9 how it works4 the edge

AI Security

16 concepts · 16 questions Study module 15 →

How AI systems get attacked through their inputs, their data and their tools, and what actually stops it.

7 basics5 how it works4 the edge

Advanced AI

23 concepts · 23 questions Study module 16 →

Reinforcement learning through to RLHF, plus the compression and systems tricks that make frontier models runnable.

4 basics10 how it works9 the edge

Research Level

11 concepts · 11 questions Study module 17 →

Moving from consuming the literature to producing it: reading, reproducing, experimenting and publishing.

5 basics5 how it works1 the edge

Concept browser

All 356 concepts, A to Z, when you know the name and just want the page. The roadmap above is the reading order; this is the index.

Ablation Studies Accuracy Activation Functions Adam AdamW Adapters Adversarial Attacks Agent Evaluation Agent Loops Agentic RAG AI Guardrails Alignment APIs Attention Autoencoders Autograd and the Computational Graph Backpropagation Bag of Words Base vs Instruct Models Batch Normalization Batch vs Real-Time Inference Benchmarking BERT Bias and Variance BLEU and ROUGE Building an Eval Set Caching Calibration Canary and Shadow Deployments Catastrophic Forgetting Chain of Thought Chunking CI/CD Citations Class Imbalance Classification CLIP Cloud Basics CNNs Computer Use Conditioning and ControlNet Confusion Matrix Constraints Context Context Compaction Context Construction Context Window Contextual Retrieval Continued Pretraining Continuous Batching Convolution Corrective and Self-RAG Cosine Similarity Cosine Similarity in Search Cosine, Dot Product and Euclidean Cost Optimization Cross-Attention Cross-Entropy Cross-Validation Curse of Dimensionality Data Augmentation Data Leakage Data Leakage in AI Systems Data Poisoning Databases Dataset Dataset Evaluation Dataset Preparation DBSCAN Decision Trees Decoder Decoder-Only Architecture Denial of Wallet Dense Retrieval Developing New Methods Diffusion Models Direct and Indirect Injection Distributed Training Docker Document Ingestion DPO Dropout Early Stopping Efficient Architectures Embedding Chunks Embedding Models Embeddings in LLMs Emergent Abilities Encoder Encoder-Decoder Ensembling Epochs, Batches and Iterations Error Recovery and Retries Excessive Agency Experiment Tracking Exploration and Exploitation F1 Score FAISS Faithfulness Fallbacks and Graceful Degradation FastAPI Feature Engineering Feature Scaling Features Feed-Forward Layers Few-Shot Prompting Filters Fine-Tuning Evaluation Flash Attention Forward Propagation Full Fine-Tuning vs PEFT Function Calling GANs GPU and CUDA GPU Training Gradient Boosting Gradient Clipping Gradient Descent Gradient Descent in Networks Gradients Graph RAG Grounded Generation Groundedness GRUs Hallucination Hallucination Testing HNSW Human Evaluation Human-in-the-Loop Hybrid RAG Hybrid Search Hyperparameter Tuning Hyperparameters Image Classification Image Generation Image Segmentation Images and Pixels Implementing Architectures from Scratch In-Context Learning Indexing Inference Inference Optimization Inpainting and Outpainting Insecure Output Handling Instruction Tuning Instructions Intersection over Union IVF Indexes Jailbreaking K-Means KNN (K-Nearest Neighbours)Knowledge Cutoff Knowledge Distillation Labels / Targets Language Modelling Latency Optimization Latent Space Layer Normalization Layer Normalization in Transformers Learning Rate Schedule Linear Regression LLM Evaluation LLM-as-Judge Logging Logistic Regression Logits LoRA Loss Function Loss Functions in Deep Learning Lost in the Middle LSTMs MAE MCP (Model Context Protocol)MDPs Membership Inference and Model Inversion Memory Metadata Filtering Missing Data Mixed Precision Training Mixture-of-Experts Mode Collapse Model Compression Model Evaluation Model Extraction Model Merging Model Registry Model Routing Model Versioning Monitoring MSE Multi-Agent Systems Multi-Head Attention Multimodal Models Multimodal RAG N-Grams Naive Bayes Named Entity Recognition Nearest-Neighbour Search Neural Networks Next-Token Prediction Non-Maximum Suppression Object Detection OCR Offline vs Online Evaluation Optimization Output Parsing and Retries Overfitting OWASP LLM Top 10 Padding Pairwise Comparison and Elo Parameters Parameters in LLMs Parent Document Retrieval Parsing PCA PEFT Perceptron Perplexity PII Protection Planning Policy Policy Gradients Pooling Positional Encoding PPO PR-AUC Pre-Norm Precision Pretraining Prompt Chaining Prompt Evaluation Prompt Injection Prompt Templates Prompt Versioning Pruning Publishing Research PyTorch Q-Learning QLoRA Quantization Query Rewriting Query, Key and Value R squared RAG Evaluation RAG vs Fine-Tuning RAG vs Long Context Random Forest Rate Limits ReAct Reading Mathematical Notation Reading Research Papers Reasoning Models Recall Recall vs Latency Receptive Field Reciprocal Rank Fusion Red Teaming Reflection and Self-Critique Regression Regression Testing for Prompts Regularization Reinforcement Learning Reinforcement Learning basics Relevance Reproducing Papers Reranking Reranking in RAG Research Ethics Residual Connections ResNet Retrieval Retrieval Metrics Reward Hacking Rewards RLHF RMSE RNNs ROC-AUC Role Prompting and Personas Rotary Embeddings Running Experiments Sample Size Sampling Strategies Sandboxing Scaled Dot-Product Attention Scaling Laws Schedulers and Sampling Steps Secure Tool Use Self-Attention Self-Consistency Self-Supervised Learning Semantic Similarity Semantic, Instance and Panoptic Segmentation Sentiment Analysis Sequence-to-Sequence Serving Frameworks SGD (Stochastic Gradient Descent)Softmax Sparse Retrieval Speculative Decoding Speech-to-Text State Static vs Contextual Embeddings Staying Current Stemming and Lemmatization Stride Structured Outputs Supervised Fine-Tuning Supervised Learning Supply Chain Security Survey Papers SVM (Support Vector Machine)Synthetic Data Generation System Prompts Temperature Text Classification Text Preprocessing Text-to-Image Text-to-Speech Text-to-Video TF-IDF The KV Cache The Transformer Block Tokenization Tokenizers Tokens Tool Calling Tool Selection Tools Top-k Sampling Top-p Sampling Tracing and Observability Training Training / Validation / Test Sets Transfer Learning U-Net Underfitting Unsupervised Learning VAEs Vanishing and Exploding Gradients Vector Databases Vector Quantization Vector Spaces Vision Transformers Vision-Language Models Watermarking and Provenance Weight Initialization Weights and Biases What is Machine Learning?When Fine-Tuning Fails Why Attention Costs O(n²)Word Embeddings Workflows XGBoost YOLO Zero-Shot Prompting