AI Engineering Distilled: The Systems Design Guideline for Software Engineers and Web Developers

By: Hashemipour, Fereydun

The short, exact guideline for designing an AI system before anyone writes code. For software engineers and web developers who can build an application around a language model and now have to make one that holds: reviewable, bounded, and measured against the requirement it exists to meet.

Every AI-enabled system has the same seams: where untrusted text enters, where a model decides, where a tool writes, where a person must stand. Most architecture drawings hide them behind a box labelled "AI". This book gives you a notation that cannot draw that box, and the small set of concepts you need to use it well.

- Seam, an AI systems notation. Sixteen elements, five views (Boundary, Context, Run, Evidence, Configuration), and twelve well-formedness rules. Small enough to learn in an afternoon, strict enough that a wrong design will not draw.
- Twelve principles, from Draw the boundaries, not the AI, and Correctness is a rate, to Configuration is the system and Verify the system, validate the requirements. Each with its rationale, its consequence, and the smell that tells you it has been violated.
- Twenty-four patterns for context, control, trust, and evidence: Approval Gate, Fallback Ladder, Calibrated Judge, Provenance Seal, Configuration Record, and their neighbors, every one drawn on the same canonical system so you can compare them side by side.
- The concept phase, end to end. Needs, operating concept, requirements, architecture, and the verification and validation plan, the left side of the systems engineering V, applied to systems whose central component is probabilistic.
- Mappings to C4, arc42, UML, BPMN, SysML v2, NIST AI RMF, and ISO/IEC 5338, so the notation fits the review room you already work in.

Who it is for:
Software engineers who inherited an AI feature and need to make it defensible. Web developers and web coders moving from a working prototype to a system that has to be right at a stated rate. Architects and reviewers who need a common vocabulary for what a model call is allowed to see and do. Readers who began with vibe coding and now own a system that has to work. It is a design book, not a coding book: there is no framework to install and no code to run, and the ideas are written to outlast the current model generation.

In the tradition of the brief guides that taught a generation of engineers to draw software: one vocabulary, one notation, one worked example, and nothing you will not use.

About the author. Fereydun Hashemipour, PhD, is a national expert in artificial intelligence who has served the public and private sectors at national scale. He is the founder and CEO of Zorost Intelligence, and through Zorost AI Lab he has designed and shipped production AI platforms in aviation intelligence, evidence synthesis, regulated manufacturing, and geopolitical analysis. He builds systems end to end and writes from that practice: the concepts in this book are the ones his teams draw before they build.

• Format: Paperback
• Language: English
• ISBN: B0HHZLB6QF
• ISBN13: 9798172333262
• Release Date: September 2026
• Publisher: Independently Published
• Length: 260 Pages
• Weight: 1.34 lbs.
• Dimensions: 0.6" x 8.5" x 11.0"