What's new

Structure and languages

• The book is now trilingual: a complete new English edition, alongside Portuguese and Spanish.
• Restructured from 14 to 16 modules: Prompt and Context Engineering, RAG and GraphRAG, AI Agents, and Integration and Interoperability are now their own chapters, instead of parts of a larger module.

Deeper content

• All 16 modules were rewritten with deeper theoretical coverage (the text grew from roughly 57,000 to 78,000 words).
• New section on tacit knowledge, explicit knowledge, and the role of ontologies in sectors with rigorous technical criteria (Module 14).
• Real cases added: the ELIZA effect, the Mata v. Avianca court case, Air Canada's chatbot, and the car sold for US$ 1 by a misconfigured chatbot.
• "In Practice" sections in every module, with free, reproducible experiments (Teachable Machine, TensorFlow Playground, Transformer Explainer, Hugging Face, Gandalf, among others).

Figures and visual resources

• Dozens of new original technical figures, including interactive comparisons (click to switch) between a base model, an instruction-tuned model, and an RLHF-aligned model.
• New visual decision guides: when to use RAG vs. GraphRAG, when to trust, verify, or intervene with an agent, and more.

References and glossary

• Expanded bibliography, with more than twenty new academic references.
• Glossary nearly doubled, with terms remapped by module.

• The book's initial release: 14 modules, in Portuguese and Spanish.
• Full coverage of LLM fundamentals, from tokens and attention to RAG, agents, and security.