The Map

The series as a dependency graph

The whole premise of the series is that the LLM stack is a dependency graph, and that nothing should get used before the intuition for it has been earned. This page is that claim, drawn. Every edge comes from the posts themselves — a line means one post actually links back to another as prior material. Find a concept you need, then walk its chain backwards until you hit ground you already know. That's your reading list.

reading orderwhat it builds onwhat builds on it
ARC 01Math & ComputationARC 02Before TransformersARC 03TokenizationARC 04TransformersARC 05Decoding & InferenceARC 06Engines & ServingARC 07TrainingARC 08EvaluationARC 09Retrieval & ContextARC 10Tools & AgentsARC 11Provider EcosystemsARC 12Governance & Trajectory1.1Vectors, Matrices, and …1.2Norms, Dot Products, an…1.3Distributions, Softmax,…1.4Cross-Entropy, KL Diver…1.5Gradients and How Machi…1.6Optimizers: Momentum, A…1.7GPUs, Floating Point, a…1.8A Short Prehistory of S…2.1Language Modeling as Ne…2.2N-gram Models and the C…2.3Word2vec and the Embedd…2.4GloVe, FastText, and th…2.5Recurrent Neural Networ…2.6LSTMs, GRUs, and Gated …2.7Seq2seq, Bahdanau Atten…3.1Unicode, Bytes, and Wha…3.2Byte-Pair Encoding from…3.3WordPiece, Unigram, and…3.4Vocabulary Size, Merge …3.5The Embedding Table and…3.6Special Tokens, Chat Te…3.7Packing, Masking, and T…4.1Self-Attention: Q, K, V…4.2Multi-Head Attention an…4.3Causal Masking and the …4.4Positional Encodings: S…4.5The Feed-Forward Block …4.6Layer Normalization: Pr…4.7Decoder-Only vs. Encode…4.8A Close Reading of 'Att…4.9Mixture-of-Experts Laye…5.1Training View vs. Infer…5.2Prefill vs. Decode: The…5.3Why One New Token Means…5.4The KV Cache from First…5.5Sampling Strategies: Te…5.6Speculative Decoding5.7Continuous Batching5.8Prefix Caching and Prom…5.9Structured and Constrai…6.1What an Inference Engin…6.2Kernel Fusion and GPU O…6.3PagedAttention and the …6.4Memory Management for L…6.5Quantization: INT8, INT…6.6Tensor, Pipeline, and E…6.7Throughput vs. Latency:…6.8Comparing Engines: vLLM…6.9Benchmarking Your Own I…7.1Pretraining Data: Mixtu…7.2Distributed Training: F…7.3Supervised Fine-Tuning7.4The RLHF Pipeline: Rewa…7.5DPO and Its Variants: S…7.6Constitutional AI and R…7.7Tool-Use Fine-Tuning7.8Long-Context Training T…7.9Synthetic Data and Dist…7.10LoRA and Parameter-Effi…8.1Loss vs. Benchmarks: Tw…8.2Eval Harnesses: lm-eval…8.3Contamination and Leaka…8.4Evaluating Reasoning8.5Tool and Agent Evaluati…8.6Human Preference Evalua…8.7Calibration and Abstent…9.1Embeddings from Scratch…9.2Vector Search and Appro…9.3Dense, Sparse, and Hybr…9.4Chunking Strategies9.5Rerankers and Cross-Enc…9.6RAG Architectures End t…9.7GraphRAG and Knowledge …9.8Context Windows: Advert…9.9Context Engineering: Co…10.1A Short History of Agen…10.2Function Calling as Str…10.3The Agent Loop: Model, …10.4Transcript Formats: How…10.5MCP: A Cross-System Sta…10.6Remote MCP, OAuth, and …10.7Planning and Reasoning …10.8Computer Use and Browse…10.9Multi-Agent Systems and…11.1OpenAI's API: From Chat…11.2Anthropic's API: Messag…11.3Google's API: Gemini, G…11.4Open-Weight Models: Lla…11.5The Hugging Face Ecosys…11.6Bedrock, Azure AI, and …11.7Lock-In, Portability, a…12.1Scaling Laws: What They…12.2Inference Economics: Co…12.3Open vs. Closed: Ecosys…12.4Safety and Alignment: T…12.5The EU AI Act and Globa…12.6Standards Wars: MCP, A2…12.7A Short History of the …12.8What Comes Next: Honest…

Tap or hover any post to see what it builds on (indigo) and what builds on it (amber). A click pins it here; walk the chips backwards until you hit ground you already know. Dashed rings mark the featured posts.

Prefer a list? The series index has all twelve arcs with curated starting paths.