The map and
the one page.
Two companions to the series. The Cross-Book Navigation Guide traces one signal through three books. The Series Cheat Sheet holds the whole argument on a single page.
A series is not three books on a shelf. It is one signal traced through three systems, set inside one fictional regulated fintech, Nebula Financial. A fraud signal fires; three systems answer it, each owning a layer of one stack. NexusCore routes it, AgentMesh investigates it, ThinkFlow ships the fix.
These two resources are the reader's compass. They appear in the back matter of all three books, and they are hosted here in full so you can read them on the page or take them with you. Both are free, and both are licensed CC BY-NC-ND 4.0.
Cross-Book Navigation Guide
The chapter-to-chapter map. Where a thread picked up in one book is continued in another, organized by reading order, by source book, and by the nine research anchors.
Print-ready · CC BY-NC-ND 4.0
Series Cheat Sheet
One page. Three books, three reference architectures, nine research anchors, the incident spine, and the governance spine. Designed to be torn out, pinned up, and argued with.
Print-ready · CC BY-NC-ND 4.0
Series Cheat Sheet.
Production is not deployment. It is the architecture of trust under load.
Three books. One company. One discipline observed from three altitudes. The series is set inside Nebula Financial, a regulated fintech that runs three systems, each owning a layer of one stack. A single fraud signal traces through all three and proves why each layer must exist: NexusCore routes it, AgentMesh investigates it, ThinkFlow ships the fix.
| Book | System | Layer | Primary reader | What it teaches | |
|---|---|---|---|---|---|
| Book 1 | LLM Systems in Production | NexusCore | Infrastructure / SRE | SREs, cloud architects, platform engineers entering AI | How a request finds the right model under a latency, cost, and risk budget, and is logged as evidence. |
| Book 2 | Prompt Systems and Agent Orchestration | AgentMesh | Application / Agents | AI engineers, application developers, agent builders | How a pile of prompts becomes governed agents that plan, act, verify, and submit to human review. |
| Book 3 | DevOps for AI-Native Platforms | ThinkFlow | Operations / Platform | DevOps and platform teams, internal-platform owners | How a platform builds, ships, and governs the models, agents, and code the other two layers depend on. |
Each book stands alone. Read together, they compose one competence: the ability to design, build, and operate AI systems that survive contact with regulation, scale, and time.
The routing and observability gateway
An SLO-first control plane between every Nebula application and the pool of models behind it. It resolves into four planes.
- Data plane. Gateway to Routing Brain to Latency Controller to the governed model pool (small on-premise, mid-tier cloud, frontier) and back.
- Control plane. The Routing Brain chooses the smallest model a request can trust. The Latency Controller attaches a service objective and earns it with speculative decoding.
- Governance plane. Routing and speculation policies are signed, versioned artifacts promoted through GitOps with a conformance gate.
- Evidence plane. Every routing decision, decode strategy, model version, and cost recorded as a queryable record.
The gateway is the place an institution decides what it is allowed to think, and proves afterward that it thought it lawfully.
The catalog of record and orchestration engine
The system that turns prompts into a governed agent ecosystem for know-your-customer checks, fraud investigation, disclosure drafting, and analytics.
- Taxonomy and contracts. Agents classified by domain, capability, and autonomy, each declaring a capability contract.
- Four-module loop. Planner decomposes, executor calls tools under contract, verifier checks output, generator produces the response.
- Graph orchestration. Workflows modeled as activity-on-vertex graphs that can be inspected, parallelized, and refined.
- Human review as runtime. Tiered review (auto-approve, lightweight, full supervisory) mapped to risk, with reviewer feedback captured as first-class data.
An agent is not a clever prompt. It is a governed actor with a contract, a boundary, and a record.
The AI-native internal developer platform
The platform where Nebula builds and governs itself, extending the catalog to models and agents.
- Catalog and golden paths. Services, models, and agents as first-class catalog citizens, with paved roads for AI workloads.
- Policy-bounded delivery. Agentic gates inside the pipeline, constrained by Delivery Guardrails and a trust-tier model, every decision logged.
- Reinforcement-learned testing. An adaptive agent chooses which tests to run, skip, or parallelize, bounded by per-service risk profiles.
- PARA operations. DevOps agents structured as perception, action, reasoning, and reflection, each evaluated against benchmarks of real failures.
A platform is the paved road that decides what an organization can build without asking permission, and what it cannot build without losing control.
Three concepts per book, drawn from 2024 to 2026 research and translated into buildable patterns inside Nebula Financial. They are positioned as chapters, not footnotes. Full citations live in each book's inventory and bibliography.
| # | Book | Anchor | Home chapter |
|---|---|---|---|
| 1 | NexusCore | Universal, workload-aware LLM routing (learned routers over a model pool) | Book 1, Ch. 4 |
| 2 | NexusCore | SLO-aware speculative and pipelined decoding as an SRE primitive | Book 1, Ch. 6 |
| 3 | NexusCore | Secure and auditable router lifecycle (the router as an attack surface) | Book 1, Ch. 9 |
| 4 | AgentMesh | Trainable planner-executor-verifier-generator loops | Book 2, Ch. 3 |
| 5 | AgentMesh | Graph-based agent workflows with dynamic refinement | Book 2, Ch. 5 |
| 6 | AgentMesh | Tiered human-in-the-loop orchestration as a first-class runtime | Book 2, Ch. 8 |
| 7 | ThinkFlow | Policy-bounded AI-augmented CI/CD with trust tiers | Book 3, Ch. 4 |
| 8 | ThinkFlow | Reinforcement-learned adaptive test selection | Book 3, Ch. 6 |
| 9 | ThinkFlow | Perception-action-reasoning-reflection DevOps agents with benchmark evaluation | Book 3, Ch. 7 & 8 |
Follow the fraud signal across the series and you have read the argument the books were built to prove.
- Book 1, Ch. 4. The router selects a model worthy of the risk and logs the decision.
- Book 1, Ch. 12. The routed request is handed forward as a workflow.
- Book 2, Ch. 3. The workflow is decomposed across planner, executor, verifier, generator.
- Book 2, Ch. 8 & 11. Tiered human review engages and the workflow produces an audit record.
- Book 3, Ch. 11. The gap the investigation exposed becomes a new service the platform scaffolds, tests, and ships.
Two enterprise frameworks carry through every system as structure, not decoration. TOGAF for architecture governance. DMBOK for data governance. Routing policies become governed artifacts with provenance and promotion. Agent catalogs become managed assets with lineage. Delivery decisions become audited events. Every book's Appendix E carries the TOGAF, DMBOK, and EU AI Act / NIST AI RMF alignment checklists.
Governance is not compliance. It is coherence made visible.
Cross-Book Navigation Guide.
A series is not three books on a shelf. It is one signal traced through three systems, and this guide is the map of the thread.
Each book in this series stands on its own. A reader can finish one and be served. The reader who wants the whole competence, though, needs to know where a thread picked up in one book is continued in another. This guide is that map. It is anchored on the single incident that runs through all three systems, and it gives concrete chapter-to-chapter links so that a reader of any one book knows exactly where the argument resumes in the others.
A fraud signal fires inside Nebula Financial. Three systems answer it, each owning a layer of the same stack.
The series spine · One signal, three layers, the same discipline observed from three altitudes. The model never changes. The vantage point does.
Three reading patterns are common. Find yours, then use the chapter tables in the full guide to move between books with intent rather than by guesswork.
| If you are | Read in this order | Why |
|---|---|---|
| An SRE or platform engineer entering AI | Book 1 → Book 3 → Book 2 | Begin where reliability lives. NexusCore teaches the gateway, ThinkFlow teaches the platform that ships it, AgentMesh fills in the application layer the platform serves. |
| An agent or application builder | Book 2 → Book 1 → Book 3 | Begin where your work lives. AgentMesh teaches orchestration first, then drop down to the routing layer it sits on, then rise to the platform that ships it. |
| A platform or DevOps owner | Book 3 → Book 1 → Book 2 | Begin with the paved road. ThinkFlow teaches the platform, then NexusCore explains the model traffic it governs, then AgentMesh explains the agents it catalogs. |
This is the load-bearing path through the series. Read these five chapters across the three books and you have followed the fraud signal from the routing decision to the fix that closes the gap it exposed.
| Stage | Book and chapter | What happens to the signal |
|---|---|---|
| 1. Finds its model | Book 1, Ch. 4 | A learned router weighs prompt features, domain tags, and risk, then selects the smallest model the high-stakes path can trust. The decision is logged. |
| 2. Becomes a workflow | Book 1, Ch. 12.3 | The routed request is handed forward. A single answer becomes a case that needs decomposition. |
| 3. Is decomposed | Book 2, Ch. 3 | The investigation is broken into steps. A planner plans, executors call tools, a verifier checks the output before it leaves the system. |
| 4. Human review engages | Book 2, Ch. 8 & 11 | The case crosses a regulatory threshold, so tiered review engages and the workflow produces an audit record. |
| 5. Gap becomes a service | Book 3, Ch. 11 | The investigation exposed a missing risk-scoring service. A team requests it in natural language, and the platform scaffolds, tests, and ships it. The loop closes. |
Five chapters, three books, one signal. Everything else in the series exists to make these five chapters possible. The full guide carries the complete cross-reference tables, organized by source book, plus the nine research anchors mapped across the series. Download the PDF above for the chapter-by-chapter detail.
Each book carries a Chapter 11 case study, and the three are written to be read as one story from three vantage points. Book 1 is the incident from the infrastructure side: a bad router version degrades trading-desk latency, and a GitOps rollback restores it. Book 2 is the investigation from the agent side: a regulatory disclosure drafted by a planner-centric, graph-based workflow under tiered review. Book 3 is the resolution from the platform side: the gap the investigation exposed is filled by a new service the platform scaffolds, tests, and ships.
The template package.
The conceptual artifacts in the series, the routing policies, capability contracts, delivery guardrails, and golden paths, are governance-grounded starting points an enterprise can implement. An engagement adapts them to your stack, your data-residency boundaries, and your compliance obligations. The methodology and architectures remain the author's, licensed for use within the system you build.
© 2026 Nabeel Khan. The Full-Stack AI Engineering Series, the Cross-Book Navigation Guide, and the Series Cheat Sheet are licensed under CC BY-NC-ND 4.0, Attribution-NonCommercial-NoDerivatives. You may share them with credit to the author; you may not sell them or distribute modified versions. The frameworks, reference architectures, and named systems (NexusCore, AgentMesh, ThinkFlow) are the intellectual property of the author.