Reference

Resources

Free material to go deeper, grouped by part and chapter. Links dated 2026-10-21 were opened and checked that day. Links marked “Not yet checked” were added later and have not been opened. All of it is free to read.

Part I · Maths

2. Algebra, functions and graphs

• Linear equations & graphs (Khan Academy, Algebra 1) video beginner
  Short lessons plus practice on slope, intercepts and plotting straight lines, the exact groundwork this chapter needs. Checked 2026-10-21.
• Functions (Khan Academy, Algebra 1) video beginner
  Builds the idea of a function as an input-to-output rule, with notation, domain and range, and graded exercises. Checked 2026-10-21.
• Equation of a Straight Line article beginner
  One-page walk through y = mx + b with an interactive graph, handy as a quick refresher on slope and intercept. Checked 2026-10-21.
• What is a Function article beginner
  Plain-language picture of functions as machines that map inputs to outputs, good before reading any ML notation. Checked 2026-10-21.

3. Exponentials and logarithms

• What's so special about Euler's number e? | Chapter 5, Essence of calculus video beginner
  Shows why e appears when a quantity grows in proportion to itself, linking exponentials to their own derivative. Checked 2026-10-21.
• Logarithm Fundamentals | Ep. 6 Lockdown live math video beginner
  A gentle live lesson treating logs as the inverse of exponentials, with the rules derived rather than memorised. Checked 2026-10-21.
• Logarithms (Khan Academy, Algebra 2) video beginner
  Practice unit covering log rules, the constant e and natural log, and solving exponential equations. Checked 2026-10-21.
• An Intuitive Guide To Exponential Functions & e article beginner
  Builds e from continuous compound growth so the number feels natural instead of arbitrary. Checked 2026-10-21.

4. Vectors and matrices

• Essence of linear algebra (3Blue1Brown playlist) video beginner
  The standard visual series: vectors, linear maps and matrices as geometry, ideal before seeing them in neural nets. Checked 2026-10-21.
• Matrix multiplication as composition | Chapter 4, Essence of linear algebra video beginner
  Explains why matrix products work the way they do by chaining transformations, which is what network layers do. Checked 2026-10-21.
• Dot products and duality | Chapter 9, Essence of linear algebra video intermediate
  Gives the geometric meaning of the dot product, the operation behind similarity scores and attention. Checked 2026-10-21.
• Linear algebra (Khan Academy) video intermediate
  Full course on vectors, spaces and matrix transformations with exercises, for readers who want drill after the visuals. Checked 2026-10-21.

5. Derivatives and gradients

• Essence of calculus (3Blue1Brown playlist) video beginner
  Visual introduction to derivatives as rates of change, building intuition before any formula work. Checked 2026-10-21.
• Visualizing the chain rule and product rule | Chapter 4, Essence of calculus video beginner
  Makes the chain rule concrete with nudges flowing through nested functions, the core of backpropagation. Checked 2026-10-21.
• Multivariable calculus (Khan Academy) video intermediate
  Course whose derivatives unit covers partial derivatives and gradients for functions of many inputs. Checked 2026-10-21.
• The gradient vector (Khan Academy article) article intermediate
  Defines the gradient as the vector of partial derivatives with worked 2D and 3D examples. Checked 2026-10-21.

6. Gradient descent

• Gradient descent, how neural networks learn | Deep Learning Chapter 2 video beginner
  Pictures learning as walking downhill on a cost surface, tying the gradient to how weights get updated. Checked 2026-10-21.
• Gradient Descent, Step-by-Step video beginner
  Works a tiny regression by hand so every step size and update is visible, including the learning rate. Checked 2026-10-21.
• Linear regression: Gradient descent article beginner
  Short crash-course page on iterating toward lower loss and reading loss curves for convergence. Checked 2026-10-21.
• Linear regression: Hyperparameters article beginner
  Covers learning rate, batch size and epochs, with what happens when the learning rate is too high or too low. Checked 2026-10-21.

7. Probability and Bayes

• Bayes theorem, the geometry of changing beliefs video beginner
  Derives Bayes' rule from a picture of proportions, making updating on evidence feel obvious. Checked 2026-10-21.
• Conditional Probabilities, Clearly Explained!!! video beginner
  Slow, friendly treatment of conditional probability, the piece most people trip on before Bayes. Checked 2026-10-21.
• Seeing Theory article beginner
  Interactive visual book on probability basics, distributions and Bayesian inference you can click through. Checked 2026-10-21.
• Probability (Khan Academy, Statistics and probability) video beginner
  Practice unit on sample spaces, addition and multiplication rules, and independence. Checked 2026-10-21.

8. Mean, variance and normalisation

• Calculating the Mean, Variance and Standard Deviation, Clearly Explained!!! video beginner
  Computes the three summary numbers step by step and explains what each one tells you. Checked 2026-10-21.
• Summarizing quantitative data (Khan Academy) video beginner
  Unit on measures of centre and spread, including population versus sample variance. Checked 2026-10-21.
• Modeling data distributions (Khan Academy) video beginner
  Covers z-scores, percentiles and the normal distribution, the basis of standardising data. Checked 2026-10-21.
• Numerical data: Normalization article beginner
  Shows why features are rescaled before training and compares z-score, min-max and log scaling. Checked 2026-10-21.

9. Softmax, cross-entropy and KL divergence

• Visual Information Theory article intermediate
  Builds entropy, cross-entropy and KL divergence from pictures of message lengths; the clearest long-form read on the topic. Checked 2026-10-21.
• Neural Networks Part 5: ArgMax and SoftMax video beginner
  Shows how softmax turns raw scores into probabilities and why it is preferred to argmax for training. Checked 2026-10-21.
• Neural Networks Part 6: Cross Entropy video beginner
  Explains cross-entropy as the loss paired with softmax, with a worked numeric example. Checked 2026-10-21.
• The Key Equation Behind Probability video intermediate
  Artem Kirsanov connects surprise, entropy, cross-entropy and KL divergence and why minimising cross-entropy trains models. Checked 2026-10-21.

Part II · ML foundations

10. What learning means: linear and logistic regression

• Linear Regression, Clearly Explained!!! video beginner
  Fitting a line by least squares, R-squared and p-values, explained without heavy maths. Checked 2026-10-21.
• Linear regression: Loss article beginner
  Introduces loss functions such as squared and absolute error and how the choice changes the fitted model. Checked 2026-10-21.
• StatQuest: Logistic Regression video beginner
  Moves from predicting numbers to predicting probabilities of a class with the S-shaped curve. Checked 2026-10-21.
• Logistic regression: Loss and regularization article beginner
  Explains log loss for classifiers and why regularisation matters for logistic models. Checked 2026-10-21.

11. Generalisation, overfitting and regularisation

• Datasets, generalization, and overfitting article beginner
  Crash-course module on what generalising means and how to spot a model that memorised its data. Checked 2026-10-21.
• Datasets: Dividing the original dataset article beginner
  Why data is split into training, validation and test sets, and the leaks to avoid when splitting. Checked 2026-10-21.
• Machine Learning Fundamentals: Bias and Variance video beginner
  Short visual explanation of the bias-variance trade-off with simple and wiggly models. Checked 2026-10-21.
• Overfitting: L2 regularization article beginner
  Shows how penalising large weights reduces overfitting and how the regularisation rate is tuned. Checked 2026-10-21.

12. Neural networks and backpropagation

• Neural networks (3Blue1Brown playlist) video beginner
  The well-known visual series on neurons, layers, gradient descent, backpropagation and onward to transformers. Checked 2026-10-21.
• Backpropagation calculus | Deep Learning Chapter 4 video intermediate
  The chain-rule maths behind backprop, written out for a tiny network so each term is traceable. Checked 2026-10-21.
• The spelled-out intro to neural networks and backpropagation: building micrograd video intermediate
  Karpathy codes an autograd engine from scratch, so backprop becomes code you wrote rather than a formula. Checked 2026-10-21.
• Neural networks: Activation functions article beginner
  Why non-linear activations are needed and how sigmoid, tanh and ReLU differ. Checked 2026-10-21.

13. Training deep networks well

• An overview of gradient descent optimization algorithms article intermediate
  Readable survey of SGD, momentum, RMSprop and Adam with the update rules side by side. Checked 2026-10-21.
• Building makemore Part 3: Activations & Gradients, BatchNorm video intermediate
  Karpathy diagnoses bad initialisation and dead activations in code and motivates batch normalisation. Checked 2026-10-21.
• Adam: A Method for Stochastic Optimization (2014) paper advanced
  The original Adam optimiser paper, still the default optimiser for most deep models. Checked 2026-10-21.
• Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (2015) paper advanced
  Introduces the He (Kaiming) initialisation scheme suited to ReLU networks. Checked 2026-10-21.
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift (2015) paper advanced
  The batch norm paper: normalising activations per mini-batch to speed up and stabilise training. Checked 2026-10-21.
• Layer Normalization (2016) paper advanced
  Normalises across features instead of the batch; the variant transformers use. Checked 2026-10-21.
• Dropout: A Simple Way to Prevent Neural Networks from Overfitting (2014) paper advanced
  JMLR page for the dropout paper, free to read, on randomly dropping units as a regulariser. Checked 2026-10-21.
• Deep Residual Learning for Image Recognition (2015) paper advanced
  The ResNet paper introducing skip connections, which made very deep networks trainable. Checked 2026-10-21.

14. From perceptron to transformer: the bottlenecks

• Perceptron (Wikipedia) article beginner
  History and mechanics of the perceptron, including the XOR limitation that stalled early neural network research. Checked 2026-10-21.
• Why are deep neural networks hard to train? (Neural Networks and Deep Learning, chapter 5) article intermediate
  Michael Nielsen's free chapter showing the vanishing gradient problem with experiments and a derivation. Checked 2026-10-21.
• Deep Sparse Rectifier Neural Networks (2011) paper advanced
  Early paper arguing for ReLU units over sigmoid and tanh in deep networks. Checked 2026-10-21.
• The Unreasonable Effectiveness of Recurrent Neural Networks article intermediate
  Karpathy's classic post on what character-level RNNs can learn, a snapshot of the pre-transformer era. Checked 2026-10-21.
• Understanding LSTM Networks article intermediate
  Diagram-led explanation of LSTM gates and why they help RNNs keep long-range information. Checked 2026-10-21.
• Neural Machine Translation by Jointly Learning to Align and Translate (2014) paper advanced
  The Bahdanau attention paper, where a decoder learns where to look in the source sentence. Checked 2026-10-21.
• Attention? Attention! article intermediate
  Lilian Weng traces attention from seq2seq models through self-attention to the transformer. Checked 2026-10-21.

15. Project: train your own neural network

• The spelled-out intro to neural networks and backpropagation: building micrograd video intermediate
  Step-by-step build of a tiny autograd engine and neural net in plain Python, the template for this chapter's project. Checked 2026-10-21.
• karpathy/micrograd docs intermediate
  The finished code from the video, short enough to read in one sitting and compare against your own. Checked 2026-10-21.
• Building a neural network FROM SCRATCH (no Tensorflow/Pytorch, just numpy & math) video beginner
  Samson Zhang builds an MNIST digit classifier with only NumPy, covering forward pass, backprop and updates. Checked 2026-10-21.
• Using neural nets to recognize handwritten digits (Neural Networks and Deep Learning, chapter 1) article beginner
  Free book chapter that builds and trains a small digit-recognition network in Python with NumPy. Checked 2026-10-21.
• CS231n Deep Learning for Computer Vision article intermediate
  Stanford course notes for a minimal case study: a softmax classifier, then a two-layer NumPy network on toy data. Checked 2026-10-21.

Part III · How LLMs work

16. Language models, tokens and sampling

• Let's build the GPT Tokenizer video intermediate
  Karpathy implements byte-pair encoding from scratch and shows the odd model behaviours tokenisation causes. Checked 2026-10-21.
• Neural Machine Translation of Rare Words with Subword Units (2015) paper advanced
  The Sennrich et al. paper that brought byte-pair encoding to neural language models. Checked 2026-10-21.
• Byte-Pair Encoding tokenization (Hugging Face LLM Course) article intermediate
  Course section that trains a toy BPE tokenizer in code, with merges shown at each step. Checked 2026-10-21.
• How to generate text: using different decoding methods for language generation with Transformers article intermediate
  Compares greedy, beam, top-k and top-p sampling and shows what temperature does to the output. Checked 2026-10-21.

17. The embedding layer and positions

• Transformers, the tech behind LLMs | Deep Learning Chapter 5 video beginner
  Visual tour of the embedding layer and how token vectors carry meaning into a transformer. Checked 2026-10-21.
• Transformer Architecture: The Positional Encoding article intermediate
  Explains the sinusoidal position encoding from the original transformer and why it works. Checked 2026-10-21.
• You could have designed state of the art positional encoding article intermediate
  Walks from naive position schemes step by step to rotary embeddings (RoPE). Checked 2026-10-21.
• RoFormer: Enhanced Transformer with Rotary Position Embedding (2021) paper advanced
  The paper that introduced RoPE, now used by most open LLMs. Checked 2026-10-21.

18. The transformer block

• Attention Is All You Need (2017) paper advanced
  The original transformer paper: scaled dot-product attention, multi-head attention and the encoder-decoder stack. Checked 2026-10-21.
• The Illustrated Transformer article beginner
  Jay Alammar's diagram-by-diagram walk through self-attention and the transformer block. Checked 2026-10-21.
• Attention in transformers, step-by-step | Deep Learning Chapter 6 video beginner
  Animated explanation of queries, keys, values and multiple heads. Checked 2026-10-21.
• Transformer Explainer: LLM Transformer Model Visually Explained article beginner
  Interactive in-browser GPT-2 that lets you inspect attention and next-token probabilities live. Checked 2026-10-21.
• The Annotated Transformer article advanced
  The transformer paper reimplemented line by line in PyTorch with commentary. Checked 2026-10-21.

19. How LLMs are trained

• Deep Dive into LLMs like ChatGPT video beginner
  Karpathy's long overview of the full pipeline: pretraining on next-token prediction, fine-tuning and reinforcement learning. Checked 2026-10-21.
• Scaling Laws for Neural Language Models (2020) paper advanced
  Kaplan et al. show loss falls as a power law in model size, data and compute. Checked 2026-10-21.
• Training Compute-Optimal Large Language Models (2022) paper advanced
  The Chinchilla paper: for a fixed budget, train smaller models on far more tokens. Checked 2026-10-21.
• Training language models to follow instructions with human feedback (2022) paper advanced
  The InstructGPT paper describing supervised instruction tuning followed by RLHF. Checked 2026-10-21.
• Illustrating Reinforcement Learning from Human Feedback (RLHF) article intermediate
  Plain explanation of the three RLHF stages: pretraining, reward model, and policy optimisation. Checked 2026-10-21.
• Direct Preference Optimization: Your Language Model is Secretly a Reward Model (2023) paper advanced
  DPO trains directly on preference pairs, dropping the separate reward model and RL loop. Checked 2026-10-21.
• huggingface/evaluation-guidebook docs intermediate
  Practical guide to evaluating LLMs: benchmarks, human and model judges, and common pitfalls. Checked 2026-10-21.

20. Project: train your own tiny language model

• Neural Networks: Zero To Hero article beginner
  Index page for Karpathy's full video course, useful to plan the order of the build. Checked 2026-10-21.
• The spelled-out intro to language modeling: building makemore video beginner
  Starts with a bigram character model and turns it into a small neural language model. Checked 2026-10-21.
• Let's build GPT: from scratch, in code, spelled out. video intermediate
  Builds a small GPT on Shakespeare text step by step, the core walkthrough for this chapter's project. Checked 2026-10-21.
• karpathy/nanoGPT docs intermediate
  Minimal, readable GPT training code to compare against or extend after the video. Checked 2026-10-21.

Part IV · Applied concepts

21. Prompt engineering

• Prompting best practices (Claude API docs) docs beginner
  Anthropic's single guide to clear instructions, examples, XML-tag structure, roles and thinking. Checked 2026-10-21.
• anthropics/prompt-eng-interactive-tutorial docs beginner
  Free hands-on notebook course on prompt structure, roles, examples and avoiding hallucination. Checked 2026-10-21.
• Prompt engineering | OpenAI API docs beginner
  OpenAI's guide, including message roles and how prompting differs for reasoning models. Checked 2026-10-21.
• Language Models are Few-Shot Learners (2020) paper advanced
  The GPT-3 paper that showed in-context learning from a handful of examples in the prompt. Checked 2026-10-21.
• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (2022) paper advanced
  Shows that worked reasoning steps in examples improve multi-step answers. Checked 2026-10-21.
• Prompt caching docs intermediate
  Explains why stable content should come first in a prompt so the prefix can be cached. Checked 2026-10-21.
• Mitigate jailbreaks and prompt injections docs intermediate
  Techniques for writing prompts and pipelines that resist injected instructions. Checked 2026-10-21.
• Open Source Prompt Management - Langfuse docs intermediate
  Keeping prompts as versioned templates outside code, with labels and rollback. Checked 2026-10-21.

22. Context engineering and choosing models

• Lost in the Middle: How Language Models Use Long Contexts (2023) paper advanced
  Finds models use information at the start and end of long prompts better than in the middle. Checked 2026-10-21.
• Effective context engineering for AI agents article intermediate
  Practical guidance on what to put in context, in what order, and when to compact. Checked 2026-10-21.
• Context windows (Claude API docs) docs beginner
  How the context window fills across turns, tool calls and thinking. Checked 2026-10-21.
• Compaction overview (Claude API docs) docs intermediate
  Summarising older conversation to stay within the window on long tasks. Checked 2026-10-21.
• Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference (2024) paper advanced
  How pairwise human votes become model rankings, useful for reading public model comparisons critically. Checked 2026-10-21.
• AI Model & API Providers Analysis | Artificial Analysis article beginner
  Free site comparing models on quality, speed and price. Checked 2026-10-21.
• Compare AI Models: Pricing, Context & Benchmarks | OpenRouter docs beginner
  Catalogue of models with context length and per-token prices across providers. Checked 2026-10-21.
• Models overview (Claude API docs) docs beginner
  Anthropic's current models with context windows and capabilities side by side. Checked 2026-10-21.
• Pricing (Claude API docs) docs beginner
  Per-token prices, including cache and batch discounts. Checked 2026-10-21.

23. Embeddings and similarity

• Embeddings: What they are and why they matter article beginner
  Simon Willison's plain-English intro to text embeddings and cosine similarity with practical uses. Checked 2026-10-21.
• Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks (2019) paper advanced
  The paper behind sentence-level embedding models compared with cosine similarity. Checked 2026-10-21.
• SentenceTransformers Documentation docs intermediate
  Docs for the main Python library for computing and comparing sentence embeddings. Checked 2026-10-21.
• BAAI/bge-small-en-v1.5 docs intermediate
  Model card for the BGE embedding family, with usage snippets and notes on query instructions. Checked 2026-10-21.
• Embedding API - Jina AI docs intermediate
  Jina's embeddings page and API reference, a hosted alternative to running a local model. Checked 2026-10-21.

24. Retrieval and RAG

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020) paper advanced
  Lewis et al. paper that coined RAG: retrieve passages, then condition generation on them. Checked 2026-10-21.
• Chunking Strategies for LLM Applications article intermediate
  Compares fixed-size, sentence and structure-aware chunking and how chunk size affects retrieval. Checked 2026-10-21.
• Contextual retrieval in AI systems article intermediate
  Anthropic's write-up on adding context to chunks and combining embeddings with BM25 to cut retrieval misses. Checked 2026-10-21.
• pgvector/pgvector docs intermediate
  README for vector search inside Postgres: storing vectors, distance operators, HNSW and IVFFlat indexes. Checked 2026-10-21.
• Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs (2016) paper advanced
  The HNSW paper behind most vector indexes, including pgvector's. Checked 2026-10-21.

25. Vector search internals

• Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs (2016) paper advanced
  The HNSW paper: layered proximity graphs for fast approximate search. Checked 2026-10-21.
• Hierarchical Navigable Small Worlds (HNSW) | Pinecone article intermediate
  Free illustrated explanation of how HNSW is built and searched, with FAISS code. Checked 2026-10-21.
• The Faiss library (2024) paper advanced
  Overview of FAISS and the index families it offers, including IVF and product quantisation. Checked 2026-10-21.
• facebookresearch/faiss wiki docs intermediate
  Practical wiki for choosing and tuning FAISS indexes. Checked 2026-10-21.
• pgvector/pgvector docs intermediate
  How HNSW and IVFFlat indexes are created and tuned inside Postgres. Checked 2026-10-21.
• Okapi BM25: a non-binary model (Introduction to Information Retrieval) article intermediate
  Free textbook section deriving BM25 keyword scoring with its term-frequency and length terms. Checked 2026-10-21.
• Reciprocal rank fusion | Elasticsearch Reference docs intermediate
  Explains the RRF formula for merging keyword and vector result lists, with the usual k constant. Checked 2026-10-21.
• Cross-Encoders (Sentence Transformers documentation) docs intermediate
  Why cross-encoders are more accurate but slower than bi-encoders, and how to use them to re-rank. Checked 2026-10-21.

26. Evaluating RAG

• Evaluation Measures in Information Retrieval article beginner
  Worked examples of recall@k, precision@k, MRR, MAP and nDCG with small code snippets. Checked 2026-10-21.
• Evaluation of ranked retrieval results (Introduction to Information Retrieval, ch. 8) article intermediate
  Free textbook section defining precision at k, average precision and NDCG precisely. Checked 2026-10-21.
• Ragas: Automated Evaluation of Retrieval Augmented Generation (2023) paper advanced
  Proposes reference-free RAG metrics such as faithfulness and answer and context relevance. Checked 2026-10-21.
• Ragas documentation docs intermediate
  Docs for the Ragas library, to run those metrics on your own RAG pipeline. Checked 2026-10-21.

27. The knowledge base and maintaining it

• pgvector/pgvector docs intermediate
  Its indexing and tuning sections explain index build cost and when an index needs rebuilding. Checked 2026-10-21.
• REINDEX (PostgreSQL documentation) docs intermediate
  Official reference for rebuilding indexes, including the CONCURRENTLY option to avoid blocking writes. Checked 2026-10-21.
• HNSW indexes | Supabase Docs docs intermediate
  Practical guide to HNSW on pgvector: build parameters and behaviour as rows are added. Checked 2026-10-21.
• Document Management (LlamaIndex) docs intermediate
  Shows insert, update, delete and refresh of documents in an index so stale content gets replaced. Checked 2026-10-21.

28. Tools and function calling

• Tool use with Claude docs beginner
  Anthropic's guide to defining tools with JSON schemas and handling tool calls and results. Checked 2026-10-21.
• Function calling | OpenAI API docs beginner
  OpenAI's equivalent guide, useful for comparing how providers shape tool calls. Checked 2026-10-21.
• Writing effective tools for AI agents article intermediate
  Advice on naming, scoping and describing tools, and evaluating them with real tasks. Checked 2026-10-21.
• Toolformer: Language Models Can Teach Themselves to Use Tools (2023) paper advanced
  Research showing a model can learn when to call APIs by self-labelling its training data. Checked 2026-10-21.
• Models (Pydantic Validation docs) docs beginner
  How to declare typed data models that validate and parse input, the basis of structured outputs. Checked 2026-10-21.
• Structured outputs (Claude API docs) docs intermediate
  Anthropic's guide to making responses conform to a JSON schema. Checked 2026-10-21.
• Structured model outputs | OpenAI API docs intermediate
  OpenAI's equivalent schema-constrained output guide, for comparison. Checked 2026-10-21.
• Tool calling (Genkit, Python) docs intermediate
  How to define tools in Genkit Python and let the model call them. Checked 2026-10-21.

29. Agents and agentic flow

• ReAct: Synergizing Reasoning and Acting in Language Models (2022) paper advanced
  The paper behind the think-act-observe loop most agent frameworks use. Checked 2026-10-21.
• Building effective AI agents article beginner
  Anthropic's guide to workflow patterns versus agents and when the extra complexity pays off. Checked 2026-10-21.
• Agents - Docs by LangChain docs intermediate
  Current LangChain/LangGraph docs on building agents; the old LangGraph agents page now redirects here. Checked 2026-10-21.
• LLM Powered Autonomous Agents article intermediate
  Lilian Weng's survey of planning, memory and tool use in LLM agents. Checked 2026-10-21.
• Graph API overview - Docs by LangChain docs intermediate
  Core LangGraph concepts: state, nodes, edges, reducers, Send and Command. Checked 2026-10-21.

30. MCP, basic and advanced

• What is the Model Context Protocol (MCP)? docs beginner
  Official starting page explaining what MCP is for and its building blocks. Checked 2026-10-21.
• Architecture overview - Model Context Protocol docs beginner
  Hosts, clients, servers, the data and transport layers, and the primitives each side exposes. Checked 2026-10-21.
• Specification - Model Context Protocol (2026-07-28) docs intermediate
  Front page of the latest spec revision, linking every section below. Checked 2026-10-21.
• Key Changes - Model Context Protocol (2026-07-28) docs intermediate
  What changed in the latest revision: the initialize handshake is gone, server/discover added, Roots and Sampling deprecated. Checked 2026-10-21.
• Discovery - Model Context Protocol (2026-07-28) docs intermediate
  The server/discover call that replaces the old connection lifecycle in the latest spec. Checked 2026-10-21.
• Lifecycle - Model Context Protocol (2025-11-25) docs intermediate
  The initialize-and-shutdown lifecycle from the previous revision, still what many deployed servers speak. Checked 2026-10-21.
• Transports overview - Model Context Protocol (2026-07-28) docs intermediate
  The stdio and Streamable HTTP transports and how messages travel over each. Checked 2026-10-21.
• Tools - Model Context Protocol (2026-07-28) docs intermediate
  How servers list and run tools, with input and output schemas. Checked 2026-10-21.
• Resources - Model Context Protocol (2026-07-28) docs intermediate
  How servers expose readable data by URI. Checked 2026-10-21.
• Prompts - Model Context Protocol (2026-07-28) docs intermediate
  How servers offer reusable prompt templates to clients. Checked 2026-10-21.
• Sampling - Model Context Protocol (2026-07-28) docs intermediate
  Lets a server ask the client's model for a completion; marked deprecated in this revision. Checked 2026-10-21.
• Elicitation - Model Context Protocol (2026-07-28) docs intermediate
  How a server requests extra input from the user mid-task. Checked 2026-10-21.
• Roots - Model Context Protocol (2026-07-28) docs intermediate
  How a client tells servers which folders they may work in; marked deprecated in this revision. Checked 2026-10-21.
• Authorization - Model Context Protocol (2026-07-28) docs advanced
  OAuth-based authorization for remote HTTP servers. Checked 2026-10-21.
• Security Best Practices - Model Context Protocol docs advanced
  Known attack patterns against MCP setups, such as confused deputy and token passthrough, and their mitigations. Checked 2026-10-21.
• modelcontextprotocol/python-sdk docs intermediate
  The official Python SDK for writing MCP servers and clients. Checked 2026-10-21.
• Using server-sent events - Web APIs | MDN docs beginner
  How a server streams events to the browser over one HTTP response, the usual way to stream LLM tokens. Checked 2026-10-21.

31. Skills

• Agent Skills (Claude API docs) docs beginner
  What Agent Skills are, how they load on demand, and where they can be used. Checked 2026-10-21.
• Skill authoring best practices docs intermediate
  Guidance on concise instructions, structure and testing when writing a skill. Checked 2026-10-21.
• Equipping agents with Agent Skills article intermediate
  Anthropic's post on the design idea: folders of instructions and scripts loaded only when relevant. Checked 2026-10-21.
• anthropics/skills docs intermediate
  Public repository of example skills to read and adapt. Checked 2026-10-21.
• Extend Claude with skills - Claude Code Docs docs beginner
  How skills are created and used inside Claude Code specifically. Checked 2026-10-21.

32. Subagents

• Create custom subagents - Claude Code Docs docs beginner
  Official guide to defining subagents with their own prompt, tools and context window. Checked 2026-10-21.
• How we built our multi-agent research system article intermediate
  Anthropic's account of a lead agent spawning parallel subagents, with lessons on prompting and cost. Checked 2026-10-21.
• Building effective AI agents article beginner
  Includes the orchestrator-workers pattern, where a central model splits a task and delegates the parts. Checked 2026-10-21.
• Use the graph API - Docs by LangChain docs intermediate
  How-to guide including a map-reduce section that fans work out with the Send API and merges results. Checked 2026-10-21.

33. Harness engineering

• Building effective AI agents article beginner
  Baseline patterns for agent systems and a case for keeping them simple. Checked 2026-10-21.
• Effective context engineering for AI agents article intermediate
  Treats context as a scarce budget and covers compaction, note-taking and sub-agents. Checked 2026-10-21.
• Writing effective tools for AI agents article intermediate
  How tool design and descriptions shape agent behaviour, and how to evaluate tools. Checked 2026-10-21.
• Effective harnesses for long-running agents article advanced
  Techniques for agents that work across many context windows, such as progress files and incremental steps. Checked 2026-10-21.
• How Claude Code works - Claude Code Docs docs beginner
  Describes the agent loop behind Claude Code: gathering context, acting with tools and checking results. Checked 2026-10-21.
• Configure permissions - Claude Code Docs docs intermediate
  How allow, ask and deny rules control which tools an agent may run. Checked 2026-10-21.
• Automate actions with hooks - Claude Code Docs docs intermediate
  Running your own scripts at fixed points in the agent loop, for example before a tool call. Checked 2026-10-21.
• A practical guide to building agents article beginner
  OpenAI's free web guide on when to build an agent, design foundations and guardrails. Checked 2026-10-21.

34. Guardrails

• Presidio docs intermediate
  Docs for the PII detection and anonymisation toolkit; the project moved from Microsoft's site to this address. Checked 2026-10-21.
• Overview | NVIDIA NeMo Guardrails Library Developer Guide docs intermediate
  An open-source toolkit for input, output and dialogue rails, a useful reference architecture. Checked 2026-10-21.
• How to implement LLM guardrails article intermediate
  Cookbook notebook showing input and output checks run alongside the main model call. Checked 2026-10-21.
• Introduction - Guardrails AI docs intermediate
  Docs for a validator-based guardrail library, another pattern for checking model output. Checked 2026-10-21.

35. LLM security

• 2025 Top 10 Risk & Mitigations for LLMs and Gen AI Apps (OWASP) docs beginner
  OWASP's list of the main LLM application risks, from prompt injection to unbounded consumption, with mitigations. Checked 2026-10-21.
• Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection (2023) paper advanced
  Greshake et al. show how instructions hidden in retrieved content can hijack an LLM app. Checked 2026-10-21.
• Simon Willison: Prompt injection article beginner
  Running series of posts tracking prompt injection attacks and why they remain hard to fix. Checked 2026-10-21.
• The lethal trifecta for AI agents: private data, untrusted content, and external communication article beginner
  A simple rule for spotting agent designs that can leak data through injection. Checked 2026-10-21.
• Mitigate jailbreaks and prompt injections docs intermediate
  Anthropic's practical defences: input screening, prompt structure and monitoring. Checked 2026-10-21.

36. Hallucination

• A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023) paper advanced
  Broad survey of hallucination types, causes, detection and mitigation. Checked 2026-10-21.
• TruthfulQA: Measuring How Models Mimic Human Falsehoods (2021) paper advanced
  Benchmark of questions where models tend to repeat popular misconceptions. Checked 2026-10-21.
• Why Language Models Hallucinate (2025) paper advanced
  Argues training and grading reward guessing over saying 'I don't know', so models bluff. Checked 2026-10-21.
• Why language models hallucinate article beginner
  OpenAI's short plain-language summary of the paper above. Checked 2026-10-21.
• Extrinsic Hallucinations in LLMs article intermediate
  Lilian Weng reviews causes, detection methods and fixes for ungrounded model claims. Checked 2026-10-21.
• How Not To Sort By Average Rating article beginner
  Short, famous post introducing the Wilson score interval for proportions from small samples, useful for eval pass rates. Checked 2026-10-21.

37. Memory

• MemGPT: Towards LLMs as Operating Systems (2023) paper advanced
  Treats the context window like RAM and external storage like disk, paging memories in and out. Checked 2026-10-21.
• vectorize-io/hindsight docs intermediate
  Open-source agent memory server repository with setup instructions and integrations. Checked 2026-10-21.
• Overview | Hindsight docs intermediate
  Official Hindsight docs covering concepts, API reference and best practices. Checked 2026-10-21.
• Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects (2025) paper advanced
  The paper describing Hindsight's memory design, linked from the project's README. Checked 2026-10-21.
• Memory overview - Docs by LangChain docs intermediate
  Explains short-term versus long-term memory for agents and how LangGraph stores each. Checked 2026-10-21.

38. Frameworks and model providers

• LangGraph overview - Docs by LangChain docs intermediate
  Entry point for LangGraph, the graph-based orchestration library for stateful agents. Checked 2026-10-21.
• LangChain overview - Docs by LangChain docs intermediate
  Entry point for LangChain's model, tool and agent abstractions built on LangGraph. Checked 2026-10-21.
• Genkit docs intermediate
  Google's open-source framework for AI features, with flows, tools and structured output. Checked 2026-10-21.
• Provider Routing - Smart Multi-Provider Request Management docs intermediate
  How OpenRouter picks among providers for one model and how to steer that choice per request. Checked 2026-10-21.
• Zero Data Retention - How OpenRouter gives you control over your data docs intermediate
  How to restrict requests to providers that do not retain prompts or outputs. Checked 2026-10-21.
• Prompt caching docs intermediate
  Anthropic's guide to caching stable prompt prefixes to cut cost and latency. Checked 2026-10-21.
• Generating content with AI models (Genkit, Python) docs intermediate
  Genkit Python docs; its structured-output section shows typed responses from a schema. Checked 2026-10-21.
• Streaming - Docs by LangChain docs intermediate
  The stream modes LangGraph offers for state updates, tokens and custom events. Checked 2026-10-21.

39. Quantisation

• A Visual Guide to Quantization article beginner
  Maarten Grootendorst's illustrated tour of number formats, int8/int4, GPTQ, GGUF and more. Checked 2026-10-21.
• Overview (Hugging Face Transformers quantization docs) docs intermediate
  Compares the quantisation backends Transformers supports and when to pick each. Checked 2026-10-21.
• LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale (2022) paper advanced
  Shows outlier features break naive int8 and handles them separately to keep accuracy. Checked 2026-10-21.
• GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers (2022) paper advanced
  One-shot 3 to 4-bit weight quantisation using approximate second-order information. Checked 2026-10-21.
• AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration (2023) paper advanced
  Protects the small share of weights that matter most, judged by activation size. Checked 2026-10-21.
• QLoRA: Efficient Finetuning of Quantized LLMs (2023) paper advanced
  Introduces 4-bit NF4 quantisation so LoRA fine-tuning fits on a single GPU. Checked 2026-10-21.
• ggml-org/llama.cpp docs intermediate
  C/C++ inference engine that runs quantised GGUF models on laptops, including Apple silicon. Checked 2026-10-21.
• GGUF (Hugging Face Hub docs) docs intermediate
  Explains the GGUF file format and its quantisation types as listed on the Hub. Checked 2026-10-21.

40. Fine-tuning and distillation

• LoRA: Low-Rank Adaptation of Large Language Models (2021) paper advanced
  Freezes the base model and trains small low-rank matrices, the basis of most cheap fine-tuning. Checked 2026-10-21.
• QLoRA: Efficient Finetuning of Quantized LLMs (2023) paper advanced
  Combines a 4-bit frozen base with LoRA adapters to fine-tune large models on modest hardware. Checked 2026-10-21.
• Distilling the Knowledge in a Neural Network (2015) paper advanced
  Hinton et al. train a small student on a large teacher's softened outputs. Checked 2026-10-21.
• PEFT (Hugging Face docs) docs intermediate
  Library docs for parameter-efficient methods such as LoRA, with conceptual guides. Checked 2026-10-21.
• Practical Tips for Finetuning LLMs Using LoRA (Low-Rank Adaptation) article intermediate
  Sebastian Raschka's free post on rank, alpha, which layers to adapt and memory trade-offs from many runs. Checked 2026-10-21.

41. Projects: LoRA fine-tune and distillation

• Fine-tuning (Hugging Face Transformers docs) docs intermediate
  Official walkthrough of fine-tuning a pretrained model with the Trainer API. Checked 2026-10-21.
• Quicktour (PEFT) docs intermediate
  Minimal code to wrap a model with a LoRA config, train it and save the adapter. Checked 2026-10-21.
• SFT Trainer (TRL) docs intermediate
  Supervised fine-tuning trainer for chat or instruction data, with built-in PEFT support. Checked 2026-10-21.
• HuggingFaceTB/SmolLM2-135M docs beginner
  Model card for a tiny base model that trains quickly on a laptop, a good student candidate. Checked 2026-10-21.
• Qwen/Qwen2.5-0.5B-Instruct docs beginner
  Model card for a small instruction-tuned model with chat template usage examples. Checked 2026-10-21.
• MPS backend (PyTorch docs) docs intermediate
  How to run PyTorch training on Apple-silicon GPUs via the mps device. Checked 2026-10-21.

42. Running AI in production

• vLLM docs intermediate
  Docs for a high-throughput open-source serving engine for self-hosted models. Checked 2026-10-21.
• Ollama documentation docs beginner
  Running open models locally behind a simple HTTP API. Checked 2026-10-21.
• Prompt caching docs intermediate
  Cutting cost and latency on repeated prompt prefixes. Checked 2026-10-21.
• Batch processing (Claude API docs) docs intermediate
  The Message Batches API for large asynchronous jobs at lower cost. Checked 2026-10-21.
• Streaming messages (Claude API docs) docs intermediate
  The server-sent event stream format and how to handle each event type. Checked 2026-10-21.
• Claude API errors docs intermediate
  Error types and status codes, including which are worth retrying. Checked 2026-10-21.
• Rate limits (Claude API docs) docs intermediate
  How limits are measured, the response headers to watch, and backing off. Checked 2026-10-21.
• Handling Overload (Site Reliability Engineering, chapter 21) article advanced
  Free chapter from Google's SRE book on client-side throttling, criticality and retry budgets. Checked 2026-10-21.
• Open Source Prompt Management - Langfuse docs intermediate
  Versioned prompts fetched at runtime, so prompt changes ship without redeploying. Checked 2026-10-21.
• Datasets - Langfuse docs intermediate
  Building test sets from production traces and running experiments against them. Checked 2026-10-21.

43. AI in a regulated domain

• Artificial Intelligence Risk Management Framework (AI RMF 1.0) docs intermediate
  NIST's free framework organising AI risk work into govern, map, measure and manage. Checked 2026-10-21.
• Model Cards for Model Reporting (2018) paper intermediate
  Proposes a short standard document describing a model's intended use, evaluation and limits. Checked 2026-10-21.
• Datasheets for Datasets (2018) paper intermediate
  The dataset counterpart: recording how data was collected, what it contains and its known gaps. Checked 2026-10-21.
• People + AI Guidebook docs beginner
  Google PAIR's free practical guide to designing AI features people can understand and trust. Checked 2026-10-21.
• Guidelines for Human-AI Interaction - Microsoft HAX Toolkit docs beginner
  Microsoft's research-based guidelines for how AI features should behave with users. Checked 2026-10-21.
• Principles to Promote Fairness, Ethics, Accountability and Transparency (FEAT) in the Use of Artificial Intelligence and Data Analytics in Singapore's Financial Sector article intermediate
  A financial regulator's principles for responsible AI in finance, free to read. Checked 2026-10-21.
• AI Principles Overview - OECD.AI article beginner
  The OECD's internationally agreed principles for trustworthy AI, a short shared vocabulary. Checked 2026-10-21.

44. Evaluating LLM output: judges and agreement

• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena (2023) paper advanced
  Tests how well strong models agree with human raters and lists judge biases such as position and verbosity. Checked 2026-10-21.
• Evaluating the Effectiveness of LLM-Evaluators (aka LLM-as-Judge) article intermediate
  Eugene Yan's long review of judge designs, metrics and what the research says works. Checked 2026-10-21.
• Cohen's kappa (Wikipedia) article beginner
  Definition and worked example of agreement corrected for chance, used to compare a judge with humans. Checked 2026-10-21.
• Interrater reliability: the kappa statistic article intermediate
  Free open-access paper explaining kappa, how to interpret it and its limits. Checked 2026-10-21.
• LLM-as-a-Judge - Langfuse docs intermediate
  How to set up model-based evaluators on traces and datasets in Langfuse. Checked 2026-10-21.

45. Observability, tracing and cost

• LLM Observability & Application Tracing (Open Source) - Langfuse docs intermediate
  Langfuse's tracing docs: traces, spans and generations for LLM apps. Checked 2026-10-21.
• OpenTelemetry (OTEL) for LLM Observability - Langfuse docs intermediate
  How Langfuse ingests OpenTelemetry spans, so standard instrumentation lands in the same place. Checked 2026-10-21.
• Python (OpenTelemetry docs) docs intermediate
  Official guide to instrumenting Python services with OpenTelemetry traces and metrics. Checked 2026-10-21.

Part V · Build it

Labs: tools you will use

• Claude API documentation docs beginner
  Anthropic's official docs for the Messages API, tool use, structured outputs and pricing. The labs call the API through the Python SDK. Not yet checked.
• anthropics/anthropic-sdk-python docs beginner
  Source and README of the official Python client: installing it, sending a message, streaming and handling errors. Not yet checked.
• Ollama docs beginner
  Download page and model library for running open-weight models on your own machine, so the labs can run without a paid API. Not yet checked.
• ollama/ollama docs beginner
  The README covers installing Ollama, pulling and running a model, and its local HTTP API. Not yet checked.
• Model Context Protocol docs beginner
  Home of the protocol: introduction, SDKs and tutorials for building a server, the subject of the MCP lab. Deeper pages are listed under chapter 30. Not yet checked.
• Virtual Environments and Packages (Python tutorial) docs beginner
  How to create a virtual environment and install packages into it, the first step of the setup lab. Not yet checked.
• pytest documentation docs beginner
  Official docs for the test runner the labs use to check their own code. Not yet checked.
• uv docs beginner
  Docs for the fast Python package and project manager: projects, lockfiles and running scripts. Checked 2026-10-21.