Generative AI, Large Language Models, and Coding Agents
Foundations, adaptation, evaluation, and practical Claude workflows
2026-11-15
Introduction
This book combines two formerly separate sets of study notes: foundations of generative AI and large language models, and practical workflows for working with Claude and AI coding agents. The goal is to connect model concepts with safe, reproducible practice for data science and software work.
The book has four parts:
- LLM foundations — transformers and attention, pre-training and scaling laws, fine-tuning and PEFT, evaluation, alignment (RLHF, DPO), and building LLM applications (inference optimization, RAG, tools, responsible AI).
- Prompt engineering — prompt anatomy, techniques, use-case strategies, evaluation, and safety (in progress).
- Agentic AI — vendor-neutral design of agentic workflows: reflection, tool use and code execution, evaluation and error analysis, planning, and multi-agent systems.
- Applied agent workflows — prompting, code review, verification loops, Claude Code hooks and skills, reproducible notebooks, MCP, and supervisor agents on Databricks.
The appendices walk through the hands-on labs that accompanied the LLM foundations, Agentic AI, and context engineering material.
These are educational notes, not vendor certification material. Product features change quickly, so product-specific chapters link to primary vendor documentation and should be checked against the installed product version.
How to use this book
- Read the LLM chapters in sequence when learning the underlying concepts.
- Read the Agentic AI part before building agents; the Claude chapters then serve as practical, tool-specific references.
- Run and inspect generated code; do not treat model output as validated merely because it is fluent.
- Protect secrets and private data, use least-privilege tool permissions, and keep a human reviewer responsible for consequential decisions.