Automated Textbook Auditing with Multi-Agent LLM Systems

The paper introduces AI Textbook Auditor, a modular multi-agent pipeline for auditing textbook PDFs. Its factual and technical track uses specialized LLM agents to identify factual inaccuracies, code errors, incorrect definitions, and conceptual inconsistencies, while a PDF-native grammar track preserves diacritical encoding. A Judge Agent filters false positives using domain-specific rules before human review. On two Romanian upper-secondary textbooks, the system reported 56 technical findings in a computer science textbook, with expert-validated precision of 62.5%, and 72 findings in a history and social sciences textbook, including factual errors, ideological bias, and grammar issues.