The Craft of Post-Training

The definitive guide to transforming foundation models into production-ready enterprise AI through post-training techniques

Foundation models like GPT, Claude, and Llama have become commodities—available to every organisation through APIs or open weights. Yet most enterprises struggle to derive lasting competitive advantage from AI. The missing piece is post-training: the techniques that transform generic foundation models into specialised, aligned, production-ready systems tailored to specific business needs.

This book bridges the gap between academic research and practical implementation, covering supervised fine-tuning, RLHF and modern preference optimisation methods, evaluation discipline, efficiency techniques, domain adaptation, tool use and agency, and more. Spanning 13 chapters across 4 parts, it provides both the technical depth ML engineers need for implementation and the strategic context technical leaders require for investment decisions.

Supervised Fine-Tuning

Teaching models desired behaviors through demonstration. Data curation, training dynamics, and when SFT is enough.

Reinforcement Learning from Human Feedback

RLHF, PPO, and modern preference optimisation methods including DPO, KTO, and GRPO.

Evaluation Discipline

Building robust benchmarks, model-based evaluation, and avoiding common pitfalls in measuring quality.

Efficiency Techniques

LoRA, QLoRA, quantization, and knowledge distillation for deployment at scale.

Domain Adaptation & Tool Use

Specialising models for industries and enabling interaction with external systems through function calling.

Self-Play & Multimodal

Synthetic data generation, RLAIF, vision-language alignment, and emerging frontiers.

Why this book?

Addresses the Crucial Middle

There are books on building foundation models (for research labs) and books on using APIs (for hobbyists). There's nothing for the crucial middle: enterprises with real compute budgets who need to customize, align, and deploy AI at scale. This book fills that gap.

Trade-offs, Not Best Practices

The book treats post-training decisions as trade-offs rather than best practices, helping practitioners match techniques to constraints. It provides decision frameworks for when to fine-tune vs. use prompting, RAG, or other approaches.

From Principles to Practice

Combines technical depth with strategic context. Includes 12 companion Jupyter notebooks covering practical implementation. Shows how to embed proprietary knowledge, organisational values, and domain expertise into foundation models.

About the author

Chris von Csefalvay is a Principal at HCLTech's AI practice, where he specialises in research and clinical intelligence. He brings the dual perspectives of a researcher and an enterprise AI practitioner to the task of advising the world's leading companies on AI strategy and implementation.

In a career spanning nearly two decades, he has served in senior data science leadership roles across leading enterprises, published extensively on distributed computing for ML and designed language models for applications from pharmacovigilance to social dynamics. He is also the author of Computational Modeling of Infectious Disease, a leading monograph on computational epidemiology, the author of numerous peer reviewed papers and co-creator of one of the largest curated data sets on COVID-19.

He holds undergraduate and graduate degrees from the University of Oxford and Cardiff University, and is a Fellow of the Royal Society for Public Health and Senior Member of IEEE.

Get notified when the book launches

Be the first to know when The Craft of Post-Training is available. Sign up to receive updates on the release date, early access opportunities, and exclusive content.