How We Built This Site

This intelligent textbook was created in approximately two days using Claude Code with custom skills designed specifically for generating educational content. This page documents our process, tools, and lessons learned.

Timeline

Based on our git commit history, here's how the project evolved:

Day 1: January 31, 2026 — Foundation and Content

Time Block	Activity
Morning	Created learning graph with 300 concepts and the vis-network viewer
Late Morning	Established chapter structure and added detailed tone guide to CLAUDE.md
Midday	Created Gyra, our self-balancing robot mascot, with personality specification
Afternoon	Generated chapters 1-9 content using the chapter-content-generator skill
Evening	Completed chapters 10-16, finishing all core educational content

Day 2: February 1, 2026 — Interactive Features and Polish

Time Block	Activity
Morning	Created MicroSims including the "Control Systems in Daily Life" simulation
Midday	Built the MicroSim index catalog and added disturbance rejection simulations
Afternoon	Added GLightbox for image zoom, generated glossary and FAQ
Evening	Created feature checklist, updated README, final polish

Tools and Skills Used

Claude Code Skills

We used several custom skills from the intelligent textbook ecosystem:

Skill	Purpose	Time Saved
learning-graph-generator	Created 300 concepts with dependencies	Hours → Minutes
book-chapter-generator	Designed optimal chapter structure	Hours → Minutes
chapter-content-generator	Generated 16 chapters of educational content	Weeks → Hours
glossary-generator	Created ~300 searchable definitions	Days → Minutes
faq-generator	Generated ~50-100 common student Q&As	Days → Minutes
microsim-generator	Created interactive browser-based simulations	Hours → Minutes each
book-metrics-generator	Tracked word counts and completeness	Hours → Seconds

Cover Image Generation

The skill for generating cover images and social media preview cards was particularly helpful. Having a consistent visual identity across the site—from the Gyra mascot to chapter headers—would have taken a designer days to produce manually.

Supporting Scripts Library

To minimize token usage and maximize efficiency, we developed a library of shell scripts and Python programs that handle repetitive tasks outside of Claude:

~/.local/bin/
├── bk-capture-screenshot    # Capture MicroSim screenshots with headless Chrome
├── csv-to-json.py           # Convert learning graph CSV to vis-network JSON
└── ... (other utilities)

These scripts handle tasks like:

Screenshot capture: Headless Chrome rendering of MicroSims for social cards
Format conversion: CSV to JSON for the learning graph viewer
File organization: Batch renaming and directory structure management
Validation: Checking file formats and link integrity

By offloading these mechanical tasks to scripts, Claude can focus on creative work like writing explanations and designing interactions.

Token Efficiency: A Design Goal

The Cost Reality

We developed this textbook using Anthropic's Max plan (currently around $100/month), which provides generous token limits for intensive development work. However, a key design goal is making textbook creation accessible to educators on the Pro plan ($20/month).

Strategies for Token Efficiency

Reusable Skills: Each skill encapsulates best practices, so Claude doesn't need to figure out formatting and structure from scratch each time.
CLAUDE.md Context: Project-specific instructions in CLAUDE.md mean Claude understands conventions without lengthy explanations in each prompt.
External Scripts: Python and shell scripts handle data transformation, so tokens aren't spent on routine file manipulation.
Incremental Generation: Generate one chapter at a time rather than attempting the whole book in one session.
Template Reuse: MicroSim templates mean each new simulation builds on proven patterns.

Estimated Token Usage

Task	Approximate Tokens
Learning graph (300 concepts)	~50K
Chapter structure design	~20K
Each chapter (~3,000 words)	~15K
Each MicroSim	~10K
Glossary (~300 terms)	~30K
FAQ (~100 questions)	~25K
Total for complete textbook	~400K-500K

With careful prompting and skill usage, a complete intelligent textbook can be generated within the Pro plan's monthly limits, though it may require spreading work across several weeks.

Reproducibility

All prompts used to generate this content are saved in the prompts/ directory. This means:

Content can be regenerated if improvements are needed
Other educators can adapt our prompts for their subjects
The generation process is transparent and auditable

Lessons Learned

What Worked Well

Starting with the learning graph: Defining all concepts and dependencies first created a solid foundation for everything else
Gyra as narrative anchor: Having a mascot gave abstract concepts a relatable physical interpretation
MicroSims for intuition: Interactive simulations convey understanding that static text cannot
Conversational tone: Students engage more with friendly, accessible writing

What We'd Do Differently

More MicroSims earlier: We have 6 simulations but aim for 20+; starting MicroSim development on Day 1 would have been better
Per-chapter references from the start: Adding curated references during chapter generation is more efficient than retrofitting
Quiz generation alongside content: Writing quizzes while the chapter content is fresh reduces context-switching

Try It Yourself

If you want to create your own intelligent textbook:

Start with a clear course description including learning objectives
Use the learning-graph-generator skill to map concepts
Use book-chapter-generator to design structure
Generate chapters with chapter-content-generator
Add interactivity with microsim-generator
Polish with glossary, FAQ, and quizzes

The entire process can be completed in a weekend for a motivated author with domain expertise.

Note

This page itself was generated using Claude Code on February 1, 2026 by scanning the log files in the GitHub repository. Hooks have also been used to track both time and token usage.