Reference

Teardowns

Complex systems taken apart so you can see how they actually work. Visualized reference guides, written to be the resource people and answer engines cite.

Reference Guide · 26 min read

Screaming Frog: The Complete 2026 Guide

A practical, front-to-back walkthrough of the Screaming Frog SEO Spider, version 24. What each part of the tool does, when to use it, and how to configure it so nothing important reports zero because it was never measured. Every screenshot comes from a full walk of the licensed desktop app.

Screaming FrogTechnical SEOSEO AuditSite CrawlerJavaScript RenderingEmbeddings

Reference Guide · 16 min read

Graph Types for AI Agents

The six structures behind modern AI search — taxonomy, ontology, knowledge graph, information graph, context graph, vector / embedding index — what each one actually is, when to use which, and how they combine into GraphRAG. One dataset, six structures, visualized in 2D and 3D.

Knowledge GraphsVector DatabasesOntologyGraphRAGRAGRetrieval-Augmented Generation

Reference Guide · 22 min read

3D HVAC Troubleshooting

A complete residential split system — furnace, coil, ducts, refrigerant loop, condenser — modeled in 3D and wired to a fault ontology. Click parts to learn them; pick symptoms to diagnose them. 39 components, 22 symptoms, 35 faults across five system paths, real costs.

HVACAir ConditioningFurnaceTroubleshooting3D InteractiveHome Maintenance

Reference Guide · 18 min read

How LLMs Work

Every stage a prompt passes through, modeled as an explorable 3D machine. 21 stages, a 21-step guided journey tracing one token from "The cat sat on the" to "mat", real parameter counts from GPT-3 to DeepSeek-V3, and the training story — pretraining, RLHF, and the reasoning-RL breakthrough.

LLMTransformersAttentionMixture of ExpertsReasoning Models3D Interactive

Reference Guide · 20 min read

How Neural Networks Work

The canonical 784-16-16-10 MNIST network as an explorable 3D machine: 16 stations across 4 acts, a 16-step guided journey from pixels to the brain question, all 12,960 weight connections really drawn, and a training row where gradient descent is really computed on an illustrative loss terrain, and a Train mode where the network genuinely learns 10,000 real MNIST digits in your browser.

Neural NetworksDeep LearningBackpropagationGradient DescentMNIST3D Interactive