Agentic AI-generated Book Library

Author: E.D. Gennatas
Date: 2025-04-19

This is a project to test the abilities of AI Agents to create and maintain educational content. To generate each version of each book, we start by creating a list of chapter names and short descriptions. A ReAct agent is initialized using an appropriate system prompt and is asked to generate the content of each chapter. Part of the agent's instructions include citing its sources and declaring use of tools.

The plan is to generate each book using multiple LLMs and re-generate each version over time. This allows us to observe the evolution of both the models' abilities and the content itself.

The ultimate goal is to create tools for educational content generation that:

• can be customized to account for individual learner variability
• can create a personalized learning experience for any academic pathway.

As always, generative AI can output incorrect information. Current testing is performed using smaller models of under 10 billion parameters. The markdown output in this site is 100% unaltered. This means that it will includes errors, include broken links, etc.. Output quality will increase as models are updated and as we add larger models.