Available NowAll levels

AI Mastery Map 🗺️

A complete, level-by-level map of the AI landscape. Start at Level 1 for fundamentals and progress through machine learning, deep learning, transformers, prompting, RAG, agents, customization, production engineering, advanced strategy, recommender systems, responsible AI, and AI product craft.

Levels🗺 13

Concepts📚 152

Estimated Time⏱ Self-paced

Format⚡ Self-paced Mastery Map

The Mastery Map

13 levels and 152 concepts, from fundamentals to production.

L01

AI fundamentals

Beginner10 concepts

What AI is and how it works

Artificial intelligence (AI)Machine learning vs deep learningGenerative AI vs predictive AI

Concepts (10)

1
Artificial intelligence (AI)
2
Machine learning vs deep learning
3
Generative AI vs predictive AI
4
Supervised learning
5
Unsupervised learning
6
Reinforcement learning
7
Training vs inference
8
Datasets and data quality
9
AI model parameters
10
Probability and linear algebra foundations

L02

Machine learning+4 new

Beginner14 concepts

How models are built and measured

RegressionClassificationDecision trees and random forests

Concepts (14)

11
Regression
12
Classification
13
Decision trees and random forests
14
Gradient boosting
15
Clustering
16
Feature engineering
17
Training, validation and test splits
18
Overfitting and underfitting
19
Loss functions and optimization
20
Precision, recall, F1 and ROC-AUC
21
Regularization: dropout, weight decay, early stopping
Techniques that stop a model memorizing its training data
22
Cross-entropy and perplexity
The loss LLMs are trained on and the standard measure of language-model fit
23
Dimensionality reduction: PCA, t-SNE, UMAP
Compressing and visualizing high-dimensional data such as embeddings
24
Time-series forecasting and anomaly detection
Predicting demand and spotting outliers, e.g. network load or streaming QoE drops

L03

Deep learning+3 new

Beginner13 concepts

The technology behind modern AI

Artificial neural networksNeurons, weights and biasesActivation functions

Concepts (13)

25
Artificial neural networks
26
Neurons, weights and biases
27
Activation functions
28
Backpropagation
29
Gradient descent and optimizers
30
Convolutional neural networks
31
Recurrent neural networks
32
Attention mechanisms
33
Representation learning
34
PyTorch and automatic differentiation
35
Transfer learning
Reusing a model trained on one task as the starting point for another
36
Autoencoders and VAEs
Networks that learn compressed representations; the ancestor of latent diffusion
37
Contrastive learning and CLIP
Training shared text–image embeddings by pulling matching pairs together

L04

Transformers and LLMs+2 new

Beginner12 concepts

How language models actually operate

Large language models (LLMs)Tokenization and tokensTransformer architecture

Concepts (12)

38
Large language models (LLMs)
39
Tokenization and tokens
40
Transformer architecture
41
Self-attention and multi-head attention
42
Positional encoding
43
Pretraining and next-token prediction
44
Context windows
45
Temperature and sampling
46
Model scaling and mixture-of-experts
47
Hallucinations and grounding
48
Encoder vs decoder models: BERT vs GPT
Why some models understand, some generate, and which to use for classification or search
49
Scaling laws
How loss improves with parameters, data and compute, and what that implies for cost

L05

Prompt and context engineering

Intermediate10 concepts

Control and structure model behavior

System prompts and instruction hierarchyZero-shot and few-shot promptingPrompt templates and variables

Concepts (10)

50
System prompts and instruction hierarchy
51
Zero-shot and few-shot prompting
52
Prompt templates and variables
53
Structured outputs and JSON schemas
54
Reasoning strategies and task decomposition
55
Context engineering
56
Prompt caching
57
Long-context management
58
Prompt injection and instruction security
59
Conversation memory and state management

L06

RAG and knowledge systems

Intermediate10 concepts

Connect AI to trusted enterprise knowledge

Retrieval-Augmented Generation (RAG)EmbeddingsVector databases

Concepts (10)

60
Retrieval-Augmented Generation (RAG)
61
Embeddings
62
Vector databases
63
Similarity search and distance metrics
64
Document parsing and chunking
65
Sparse, dense and hybrid retrieval
66
Reranking
67
Knowledge graphs and GraphRAG
68
Citation grounding and source attribution
69
Retrieval evaluation and freshness

L07

Agents and tool integration+3 new

Intermediate13 concepts

Build systems that take actions

AI agentsTool and function callingModel Context Protocol (MCP)

Concepts (13)

70
AI agents
71
Tool and function calling
72
Model Context Protocol (MCP)
73
Agent planning and execution loops
74
Sub-agents and orchestration
75
Agent memory and persistent state
76
Multi-agent collaboration
77
Workflow automation vs autonomous agents
78
Human-in-the-loop approvals
79
Agent guardrails and permissions
80
Computer-use and browser agents
Agents that operate screens and websites the way a person does
81
Agent-to-agent protocols (A2A)
Standards for agents from different vendors to discover and delegate to each other
82
Evaluating multi-step agent tasks
Measuring task success, trajectory quality and cost across long agent runs

L08

Model customization and evaluation+2 new

Intermediate12 concepts

Improve reliability and adapt models

Fine-tuningSupervised fine-tuning (SFT)LoRA and parameter-efficient fine-tuning

Concepts (12)

83
Fine-tuning
84
Supervised fine-tuning (SFT)
85
LoRA and parameter-efficient fine-tuning
86
Instruction tuning and preference optimization
87
Synthetic data generation
88
Knowledge distillation
89
Quantization
90
AI evaluations (evals)
91
LLM-as-a-judge and human evaluation
92
Red teaming and adversarial testing
93
RLHF and reward models
How human preference data turns a pretrained model into an assistant
94
Data labeling and curation pipelines
Annotation guidelines, quality control and dataset versioning

L09

Production AI engineering+6 new

Advanced16 concepts

Deploy scalable, secure and economical systems

AI API integration and SDKsModel serving and inference enginesGPU, CPU and NPU acceleration

Concepts (16)

95
AI API integration and SDKs
96
Model serving and inference engines
97
GPU, CPU and NPU acceleration
98
Latency, throughput and time-to-first-token
99
Batching and KV caching
100
AI observability and tracing
101
LLMOps and CI/CD
102
Inference cost and token economics
103
AI security, privacy and governance
104
Reliability, fallbacks and production SLAs
105
Speculative decoding
A small draft model proposes tokens a large model verifies, cutting latency
106
FlashAttention and memory-efficient attention
Kernel-level tricks that make long contexts affordable
107
Semantic caching
Reusing answers for queries that mean the same thing, not just identical prompts
108
Model drift and production monitoring
Detecting when inputs or quality shift after launch
109
Rate limiting, quotas and load shedding
Keeping shared AI capacity fair and stable under peak demand
110
Guardrail and safety classifier models
Separate models that screen inputs and outputs before they reach users

L10

Advanced AI and strategy+1 new

Advanced11 concepts

Architect next-generation AI products

Multimodal AISpeech recognition and text-to-speechVision-language models

Concepts (11)

111
Multimodal AI
112
Speech recognition and text-to-speech
113
Vision-language models
114
Diffusion models and generative media
115
On-device and edge AI
116
Reasoning models and test-time compute
117
Reinforcement learning for AI systems
118
Model routing and compound AI systems
119
AI product strategy and ROI
120
AI research frontiers and emerging architectures
121
State-space models: Mamba and hybrids
Non-attention architectures with linear cost in sequence length, attractive on-device

L11

Recommender systems and personalizationPriorityNew level

Advanced10 concepts

The core AI of any content platform

Collaborative filtering and matrix factorizationContent-based filteringTwo-tower retrieval models

Concepts (10)

122
Collaborative filtering and matrix factorization
Recommending from patterns in what similar users watched
123
Content-based filtering
Recommending from item metadata: genre, cast, descriptions, embeddings
124
Two-tower retrieval models
Separate user and item encoders for fast candidate retrieval at scale
125
Learning to rank
Ordering candidates by predicted engagement or value
126
Cold start
Handling new users, new titles and new devices with no history
127
Sequential and session-based recommendation
Using what someone is watching right now, not only long-term taste
128
Multi-armed bandits and exploration
Balancing proven picks against discovery, e.g. artwork and row selection
129
Recommendation metrics: NDCG, coverage, diversity
Offline ranking quality beyond accuracy, including catalog breadth
130
Household and multi-profile personalization
Shared screens, ambiguous viewers and kids' profiles
131
LLM-powered and conversational discovery
Natural-language search and 'what should I watch' assistants

L12

Responsible AI and regulationPriorityNew level

Advanced11 concepts

Build AI that is lawful, fair and trusted

EU AI Act: risk tiers and obligationsGeneral-purpose AI model obligationsGDPR and AI

Concepts (11)

132
EU AI Act: risk tiers and obligations
Prohibited, high-risk, limited-risk and minimal-risk systems and what each requires
133
General-purpose AI model obligations
Transparency, documentation and copyright duties for foundation-model providers and deployers
134
GDPR and AI
Lawful basis, profiling, automated decisions and data-subject rights
135
Bias and fairness
Measuring and mitigating unequal outcomes across user groups
136
Explainability and interpretability
Explaining why a model produced an output, for users and regulators
137
Mechanistic interpretability
Research into the internal circuits and features of neural networks
138
Model and data licensing
Open-weight licences, usage restrictions and dataset provenance
139
Copyright and training data
Rights questions around training inputs and generated outputs
140
Federated learning
Training across devices without centralizing raw user data
141
Differential privacy
Mathematical guarantees that limit what a model reveals about any individual
142
AI alignment and safety
Making capable models reliably follow intended goals and limits

L13

AI product craftPriorityNew level

Advanced10 concepts

Turn models into products people use and pay for

AI UX patternsCommunicating uncertainty and building trustOnline A/B testing of AI features

Concepts (10)

143
AI UX patterns
Suggestions, copilots, ambient AI, undo, editing and correction flows
144
Communicating uncertainty and building trust
Confidence cues, citations and graceful failure
145
Online A/B testing of AI features
Experiment design for non-deterministic, personalized features
146
Offline vs online evaluation
Why benchmark gains don't always move live product metrics
147
Feedback loops and data flywheels
Capturing usage signals that improve the product over time
148
Defining AI success metrics
Task success, adoption, retention and cost per outcome
149
Build vs buy vs partner
When to train, fine-tune, call an API or license a vendor solution
150
Open-weight vs closed models
Control, cost, hosting and capability trade-offs
151
Model selection and vendor lock-in
Portability, abstraction layers and multi-model strategy
152
AI pricing and monetization
Usage-based, tiered and bundled models, and how inference cost shapes margin