Most writing about large language models stops at the demo. This series begins the morning after, inside a regulated institution where a model call is a billing event, an agent action is an audit record, and a deployment is a change someone has to defend. Three books carry one idea through three layers: infrastructure, application, and operations.
They follow a single fictional regulated fintech, Nebula Financial, through three systems and one continuous incident. The infrastructure layer routes a suspicious transaction for analysis. The application layer investigates it with a governed workflow of agents under human review. The operations layer ships the fix through a pipeline that grants authority by degrees and records every step.
§ 01 · One curriculum
Three books. One course of study.
Read in order, they are a curriculum. Read alone, each one is complete.
The reading order is 1 to 2 to 3, and it is the order the stack is built in. Book 1 puts a governed gateway between your applications and your models. Book 2 puts governed agents on top of that gateway. Book 3 gives the whole organization one paved road for shipping both. A reader who takes them in sequence follows one decision from a routing policy to an audit record to a governed deployment without ever losing the thread.
Each book also stands on its own, with its own reference architecture, its own reader, and its own appendices. An engineer handed only the platform problem can start at Book 3 and lose nothing. The sequence is a curriculum, not a dependency.
Book 1
InfrastructureNexusCore. An objective-first routing and observability gateway. Which model may serve a request, under which latency, cost, and risk budget, and how that decision is recorded.
LLM Systems in Production
Book 2
ApplicationAgentMesh. A catalog of record and orchestration engine for specialized agents. How agents compose into a workflow that can be inspected, reviewed, and improved without becoming opaque.
Prompt Systems & Agent Orchestration
Book 3
OperationsThinkFlow. An AI-augmented internal developer platform. How models, agents, and code reach production on one governed platform that holds authority by degrees.
DevOps for AI-Native Platforms
Figure · The series spine. One fraud incident travels through all three layers. NexusCore routes the judgment, AgentMesh investigates the case, ThinkFlow ships the fix. Three records written at three altitudes compose into one account an institution can defend.
§ 02 · The three books
The set, in reading order.
Uniform trim, uniform price, uniform standard. No book in this series is the lesser one.
Book 1 · Infrastructure · NexusCore
Book 1 of 3 · Infrastructure
LLM Systems in Production
Cloud-Native Patterns for AI Engineers
Production is not deployment. It is the architecture of trust under load. Book 1 teaches the infrastructure layer the way a site reliability engineer builds any system that cannot fail quietly: with explicit objectives, measured behavior, governed change, and an audit trail that answers who routed what, where, and why. It follows Nebula Financial as it builds NexusCore, an objective-first routing and observability gateway that sits between every application and the pool of models behind it. Along the way it replaces static threshold routing with a learned and governed routing policy, treats speculative and pipelined decoding as reliability primitives rather than research curiosities, designs observability so every routing decision is explainable to an auditor, and builds an evidence store that answers by design rather than under duress. For site reliability engineers, cloud architects, and platform engineers who already run production systems and have been handed AI workloads.
An agent is not a clever prompt. It is a governed actor with a contract, a boundary, and a record. Most teams reach the same wall: agents that pass a demo and fail in production, multiplying until no one can say what they do or what they are allowed to touch. Book 2 teaches the application layer, turning a pile of prompts into a governed agent ecosystem that plans, acts, verifies, and submits to human review. Inside Nebula Financial it builds AgentMesh, a catalog of record and orchestration engine for dozens of specialized agents, and covers the planner, executor, verifier, generator pattern and why each role must stay distinct; workflows modeled as inspectable activity-on-vertex graphs an optimizer can refine; planners trained from live trajectories without retaining what they read; tiered human review matched to risk and confidence; and reviewer feedback captured as first-class data. For AI engineers, application developers, and agent builders who have outgrown single prompts and need orchestration rather than cleverness.
Building, Governing, and Scaling AI Infrastructure
A platform is not a toolset. It is the paved road that decides what an organization can build without asking permission. Book 3 teaches the operations layer: how an organization builds, ships, and governs models, agents, and code on a single internal platform without losing control of cost, security, or audit. Inside Nebula Financial it builds ThinkFlow, an AI-augmented internal developer platform, and covers the catalog of record for services, models, and agents; golden paths that make the compliant road the easy road; policy-bounded agentic delivery with guardrails and trust tiers; reinforcement-learned adaptive testing framed as a Markov decision process; the evaluation and benchmarking of DevOps agents as measurable components; and the perception, action, reasoning, adaptation operating model with GPU-aware scheduling and per-team cost attribution. For DevOps and platform teams and internal-platform owners who must give a whole organization a paved road for AI.
Uniform pricing across the set, so the reading order is a choice about learning rather than about budget.
Formats, list prices, and ISBNs for the three books in the Full-Stack AI Engineering Series
Format
List price
Book 1 ISBN
Book 2 ISBN
Book 3 ISBN
Paperback
$49.99 USD
978-1-0678317-1-4
978-1-0678317-2-1
978-1-0678317-3-8
Hardcover
$69.99 USD
978-1-0678317-5-2
978-1-0678317-6-9
978-1-0678317-7-6
Ebook
$19.99 USD
978-1-0678317-8-3
978-1-0678317-9-0
978-1-0678960-0-3
Publisher iSystematic Inc. · First edition September 2026 · Trim 6 by 9 inches · Extent 260, 268, and 262 pages · Licence Creative Commons Attribution-NonCommercial-NoDerivatives 4.0. The chapter content is identical across formats. Prices are the publisher's list prices in United States dollars; regional pricing and taxes are set by each retailer.
The books are not on sale yet. There is no order link on this page because there is nothing honest to link to until the listings go live. Leave a name and a working email below and you get one note on publication day, with the links.
§ 04 · What is inside
A reference you work from.
Patterns, not snippets. The value is the shape of the constraint, which outlives the framework that hosts it.
3 books, 36 chapters, 790 pages, with a full set of appendices, glossaries, and references.
About 240,000 words across the set, uniform in trim, typography, and standard.
106 architectural figures in one Minimalist Flat visual system, plus a shared series-spine figure.
Nine research anchors from 2024 to 2026, three per book, translated from paper into buildable patterns.
Conceptual specifications throughout: routing policies, capability contracts, delivery guardrails, and golden-path templates, expressed as schemas and resource definitions rather than copy-paste code.
A companion repository of machine-readable specifications and editable diagram sources, opening to the public at release.
The architecture of the series is also stated on its own, in a pattern-language paper written for readers who want the academic version. It is published here in full, with the complete PDF and its references. A journal version is prepared and not yet submitted.
The reliability and software-delivery mindset is assumed. The AI-specific failure modes are new, and the series teaches them.
Book 1 is for site reliability engineers, cloud architects, and platform engineers. Book 2 is for AI engineers, application developers, and agent builders. Book 3 is for DevOps and platform teams and internal-developer-platform owners. The whole series is for architects and technical leaders who own the full stack, and for the auditors and risk officers who have to sign off on what it produces.
Site reliability engineersCloud architectsPlatform engineersAI engineersAgent buildersDevOps teamsInternal platform ownersHeads of AICTOs & VPs of EngineeringModel risk & audit
The governance invariant
A boundary is a clause the optimizer may not cross. Everything else is optimization.
One line organizes the entire series. An optimization decision improves an objective: lower latency, lower cost, higher quality, fewer tests run. A boundary decision fixes a constraint the optimizer is not permitted to relax for any gain: a residency rule, a least-privilege scope, a human-review threshold, an evidence record that must exist. A platform raises its autonomy by widening what the optimizer may do, never by relaxing a boundary. That is how it scales and stays accountable at the same time.
§ 06 · Why this series
Governance built in, not bolted on.
Most AI architecture writing rediscovers microservices and renames it. This series does something narrower and more useful. It states one invariant, the separation of a boundary the optimizer may not cross from the optimization inside it, and applies it consistently from routing to agents to delivery. Governance is part of the architecture rather than a remediation after the first incident.
The specifications are patterns, so they outlive the frameworks that happen to host them. And the three books are a single story rather than three surveys, which is why a reader can follow one decision from a routing policy to an audit record to a governed deployment without losing the thread.
NexusCore, AgentMesh, and ThinkFlow are reference architectures taught through a deliberately fictional institution. They are not the names of deployed client products. The fiction exists so the full end-to-end method can be shown without exposing any real client.
§ 07 · The author
Written by a practitioner.
NK
Author
Nabeel K.
Nabeel A. Khan is an enterprise architect and AI-governance practitioner with more than twenty-five years building systems that survive contact with regulation, scale, and time. He holds a PhD spanning neuro-marketing and computer science, and his work joins two disciplines that rarely sit together: how institutions decide, and how machines decide for them.
He writes and consults through iSystematic and grounds his teaching in systems he has built and run. He works across North America and the GCC, from stations in Toronto, Calgary, and Winnipeg, Canada, with active advisory engagements in the UAE, Qatar, Kuwait, and Saudi Arabia.
A note on cookies
This site uses Google Analytics and the Meta pixel to understand what gets read and which work reaches people. Neither one loads, and no cookie is set, until you choose. Details in the privacy & cookies notice.
