Skip to content

How This Book Was Made

This page explains who made Roots & Rivers, what AI agents and scripts did, and how each chapter was checked. A book about real people, faiths and history should be open about that.

Who did what

Aaryav Sharma, a student at Eastview High School and a member of the board of its South Asian Student Association, created the book. That meant deciding what it should be and setting its rules:

  • who it is for (high school students, both South Asian heritage teens and their friends) and what it should cover (all eight countries, not only India)
  • the structure: 22 chapters in four parts, the order of topics, and the 368 concepts in the learning graph
  • the features: the Ask Your Elders interview guides, the Whose Story? boxes, Zara and Ben, Mora the peacock, the MicroSims
  • the rules every chapter had to follow: how to handle Partition, caste, religion and disputed borders; no ranking of faiths or countries; no invented facts; every number sourced
  • reviewing the results and deciding what to publish

AI agents (Claude, made by Anthropic, running Dan McCreary's open intelligent textbook framework) drafted the chapters, built the MicroSims, and did most of the checking described below, following those written rules. Scripts counted and checked the things that can be counted: structure, links, the learning graph, and whether any fact changed during editing.

The five checking layers

  1. Sourced drafting. Each chapter's writer logged every number, date and name with the page it came from. Anything that couldn't be confirmed was cut or softened.
  2. Structure checks. Scripts confirmed that every chapter has its required sections, every assigned concept, a balanced 10-question quiz, 8 to 10 references and a working simulation.
  3. Independent fact-check. A separate agent re-checked each chapter's claims, quiz answers, links and simulation text against sources.
  4. Expert-standard edit. Another pass, written to the standard of a scholar of South Asia, looked for oversimplifications, one-sided framing and out-of-date facts, and updated them to October 2026.
  5. Plain-language edit with a fact guard. A final edit made the writing plainer and more human. A script compared every file before and after to prove that no number, name, link or fact changed.

Version 1.1 added a sixth pass: a full read of every chapter by an agent acting as a senior specialist and book editor, with every shaky claim checked again. All corrections are listed on the Accuracy and Corrections page.

What this process can't do

AI agents and scripts can check facts against sources, but they are not a substitute for a scholar who has spent a career on a subject, or for a member of a community describing their own traditions. That is why the book asks for expert reviewers, keeps a public corrections log, and builds its most important feature, Ask Your Elders, around real people. The places where a human expert should look first are listed on the Accuracy page.

Tools and sources

  • Site: MkDocs with the Material theme, hosted on GitHub Pages
  • Simulations: p5.js; charts: plain SVG
  • Framework: Dan McCreary's intelligent-textbook skills (open source)
  • Sources: listed at the end of every chapter; data from the World Bank, Pew Research Center, national censuses, UNESCO and others
  • Code and text: github.com/aaryavvatts-lab/south-asian-culture-textbook