Quarex is built on questions, not answers. Every chapter contains questions as topics, each designed to open a line of inquiry rather than test knowledge. When a question is asked, the AI uses the chapter as a contextual lens to generate an answer — and that answer generates more questions.
Every piece of content in Quarex exists within a fixed hierarchy:
This is an actual chapter from the Quarex book Sojourner Truth, located on the Political Biographies shelf in the Knowledge Libraries. The chapter covers her famous 1851 speech and the tensions between abolition and women's suffrage.
When a user selects question 2 — "Why do historians debate the famous 'Ain't I a Woman?' speech?" — the AI receives the question along with the chapter context (the chapter name, its tags, and the book it belongs to). It generates an answer that explains:
The speech exists in two versions. Marius Robinson published a contemporaneous account in 1851 that doesn't include the phrase "Ain't I a Woman?" Frances Dana Gage published a different version twelve years later, in 1863, that includes the famous refrain and renders Truth speaking in a Southern dialect — despite the fact that Truth grew up speaking Dutch in New York. Historians debate which version is authentic, or whether either is.
The AI then generates followup questions based on this answer. These might include: "How did Frances Dana Gage's version serve the political needs of 1863?" or "What other historical speeches have been retroactively altered?" Each followup opens a new path. This is recursive discovery.
AI models improve. Facts get updated. Interpretations shift. But a well-structured question — "Why do historians debate the famous 'Ain't I a Woman?' speech?" — remains valid regardless of what answer the current model produces. The questions endure. The answers are always regenerated.
The questions in a chapter define the boundaries of what that chapter covers. They are not random prompts — they are a curated argument about what matters in this topic. The questions themselves are the curriculum.
A question like "What conflicts arose between her and white feminist leaders?" does not contain a political position. It opens an area of inquiry. The AI may produce different answers over time, but the question itself is neutral by construction.
Under the hood, every chapter is stored as JSON. Here is the actual data structure for the chapter shown above:
{
"name": "Ain't I a Woman and Women's Rights (1850-1860)",
"topics": [
"What did she actually say at the 1851 Akron convention?",
"Why do historians debate the famous 'Ain't I a Woman?' speech?",
"How did she navigate tensions between abolition and women's suffrage?",
"What conflicts arose between her and white feminist leaders?",
"How did she challenge both racism and sexism simultaneously?"
],
"tags": ["politics", "activism", "gender", "civil-rights"]
}
The tags — politics, activism, gender, civil-rights — are drawn from a controlled vocabulary of 328 tags. These tags connect this chapter to every other chapter in the system that shares the same tags, enabling cross-library discovery.
Read the full Sojourner Truth book on Quarex →