NeuralyAI Field Manual · Vol. 1 · early access 2027480–520 pp · 35 chapters · 17 labs
NEURALYAI FIELD MANUALVOL. 1
FIELDCRAFT
Forward Deployed Engineering for Shipping AI Into Production: Customer Discovery, LLM Apps, GPU Inference, and the FDE Career
The whole deployment, from discovery call to GPU.
A field manual and a companion simulator for engineers who ship AI inside a customer’s environment. Find the real problem on the discovery call, build the LLM application and its evals, pass the security review, then size and run the model on the customer’s own GPUs.
+ NeuralyAI Groundplane, the Fieldcraft deployment simulator
01 · Who it’s for
For engineers who ship AI inside someone else’s building
The skill is shipping AI into a customer’s environment. “FDE” is the job title that names it most often right now.
Primary2–6 yrs
Software engineers moving into FDE work
You ship software for a living, and there is an FDE or solutions-engineering posting open in another tab. You want the whole job in one place, not ten blog posts.
The role end to end, plus the interview loop (Ch 33) and a portfolio-grade capstone.
PrimaryYear one
New FDEs in their first year
First customer, first escalation, first security questionnaire. You need to know what good looks like before Friday.
Field procedures, checklists and deliverable templates you can use this week.
SecondaryInfra
AI infra engineers and GPU-cloud architects
You know the hardware. The customer side of the job (discovery, scope, trust, comms) is the gap.
Capacity and latency maths in the customer’s context, and the 3D hardware labs.
Teams3+ hires
FDE leads onboarding a cohort
You are hiring several FDEs and want a shared vocabulary and a standard onboarding path.
Team pack: 5 seats for $399, or 10 for $699, on one invoice.
Not for you if you want a GPU-architecture reference (that is Vol. 2, AI Infrastructure) or a collection of prompt tricks.
02 · What’s inside
Eight Parts, discovery call to GPU
35 chapters, about 480–520 pages. Every chapter has a learning objective and a worked deliverable for the same customer, so each one builds on the last.
PART I · 34 PP
The Role
What the job is, how forward deployment makes money, and what a field week looks like.
01What a Forward Deployed Engineer is (and is not)
02The business of forward deployment
03The field week
PART II · 52 PP
Thinking Like an FDE
Break a vague mandate into a tree, find the real problem, scope a pilot that can be judged, and say no.
04Decomposition
05Discovery
06Scoping, success metrics and the pilot contract
07Prioritization across accounts and saying no
PART III · 62 PP
Field Engineering Foundations
Ship fast in unfamiliar code, integrate with real systems, model messy data, and deploy into someone else’s cloud.
08Shipping fast in unfamiliar territory
09Integrations with the real world
10Data integration and modeling
11Enterprise identity and access
12Deploying into someone else’s environment
PART IV · 102 PP
Building with Frontier Models
Model choice and cost, context engineering, retrieval, agents and MCP, two chapters of evals, and AI security.
13Models for engineers: selection and economics
14Prompting and context engineering
15Document AI and enterprise retrieval
16Agents, tools and MCP in production
17Evals I: eval-driven development
18Evals II: production evals and model migration
19Adapting models: fine-tune or not
20AI security: prompt injection, agent permissions and PII
PART V · 56 PP
Inference and Infrastructure
When the customer must self-host: GPU systems, memory and latency maths, Kubernetes, cluster acceptance.
21When the customer must self-host
22GPU systems in one chapter
23Capacity, latency and cost math
24Running GPU workloads on Kubernetes
25Cluster acceptance, observability and GPU triage
PART VI · 46 PP
Production, Security and Trust
Pass the security review, read the data agreements, run it in production, and hand it over.
26Passing the security review
27Legal and data agreements for engineers
28Operating in production
29Handoff, adoption and outcomes
PART VII · 34 PP
Communication and Influence
Write the update people read, run the demo and the escalation, and turn field work into product.
30Writing as an FDE
31Demos, executive communication and escalation
32Closing the loop: from field to product
PART VIII · 36 PP
Career and Sustainability
Get hired through the FDE loop, stay ethical and sane on the road, and finish the capstone.
33Getting hired
34Sustainability, travel and ethics in the field
35Growing as an FDE, and the capstone
Mission briefEach chapter opens on a week of the running engagement.
Worked deliverableA real artefact for the customer: a memo, a scope, an eval set, a runbook.
Field procedureStep by step, with the failure modes that bite in practice.
Checklist + debriefWhat to verify, and the questions a staff FDE would ask you.
03 · The running case
One engagement, start to finish
You are a Kestrune Labs FDE, four months in, assigned to Quarrybrook Health Alliance: three hospitals, 40 clinics and a health plan, with a CISO who has already blocked two AI vendors.
FROMDr. Helen Ashcombe, CMO, QHATOKestrune field engineeringREPrior authorization
Use AI to cut prior-authorization turnaround from 9 days to 2.
The whole brief, Week 0.W0 · MEMO
What you find
The decisions are not the bottleneck. About 4 in 10 requests arrive as faxed PDFs missing required clinical documents, and the review nurses spend their days chasing paperwork.
The first slice
A completeness checker that flags what is missing and cites the policy clause. A nurse makes every decision. That line is tested in Chapter 34, when an executive asks for auto-denial.
Where it goes
Hosted API under strict data terms, a security review, a board ruling that some records never leave the building, then a self-hosted model on the customer’s own GPUs.
W0The CEO memo lands. You join the account.
W1–2Decomposition and discovery on the utilization-review floor.
W2–8Fax ingestion, EHR and claims integration, SSO, deploy into their cloud.
W9Security review gate. Sensitive records must stay on-prem.
W10–16Self-hosted inference on a small GPU cluster in their datacenter.
W19Flu season: throttling GPUs and a latency SLO breach.
W24–26Handoff, the field report, and a product feature.
Quarrybrook Health Alliance, Kestrune Labs and every character are fictional. Any resemblance to real organizations or people is coincidental.
Free sample chapter
Read Chapter 4 of Fieldcraft free
Chapter 4 turns a one-line executive mandate into a tree of outcome, users, workflows, data, decisions, system and risks, then picks the first slice worth building. The full chapter as a PDF. Join the waitlist and the download link appears straight away.
Regional pricing on direct sales: up to 50% off in eligible countries.Teams: 5 seats $399 · 10 seats $699.Preview prices in USD; they may change before launch.
Not yet. It is being written now. Early access starts with v0.1 (about 80 pp) on 22 Feb 2027, followed by monthly drops; v1.0 planned for November 2027. The waitlist gets the pre-order link first.
What is a forward deployed engineer?
An engineer who works inside a customer’s environment to get a product, more and more often an AI system, into production. The job runs from the first discovery conversation through integration, deployment, evaluation and handoff. Companies use other titles for similar work, such as deployment engineer or solutions engineer.
Do I need GPU or infrastructure experience?
No. Part V starts from what a working software engineer already knows and builds up the memory, latency and cost maths step by step. Part III is skippable for experienced engineers: each chapter opens with a short skip-test.
Is Groundplane available? What does “Try the 3D lab” do?
NeuralyAI Groundplane is in development. The button tells you so and lets you join the waitlist, which also tells us how many people want the 3D lab.
The lab pages the book links to (neuralyai.com/lab/L01 to L17) are live today as free starter pages. L12 “Fit the Model” is the first lab planned to go interactive. In the simulator every number is labelled simulated and shows its formula, and it never runs your code.
Are Quarrybrook and Kestrune real companies?
No. Quarrybrook Health Alliance and Kestrune Labs are fictional, and so is every character. Field Notes in the book come only from real, consented interviews with working FDEs; anything else is labelled a composite scenario.
What formats will there be?
Early access is a DRM-free PDF and EPUB. At v1.0 the Field Kit (colour PDF, EPUB, templates) replaces it, and a 7.5 × 9.25 in paperback and a Kindle edition are planned.
Can I get a refund on early access?
Yes: refund any time before v1.0. Early access is sold on what exists, with dated drops.
Is there regional pricing?
Regional pricing on direct sales: up to 50% off in eligible countries. Amazon editions are priced per marketplace.
I’m a working FDE. How can I help?
Tick the interview box when you join. We will email you once to ask for a 30-minute call. You speak for yourself, not your employer, you never share confidential or customer information, and you approve every quote before it is used.
How is AI used in making the book?
AI tools help with research and first drafts. Every chapter is rewritten by the author, fact-checked against primary sources by a technical reviewer, and AI use is disclosed where stores require it. Stories attributed to real people are never generated.
What happens to my email?
We use it only to tell you about this book and simulator. We never sell it or share it for advertising. You can unsubscribe from any email, or ask us to delete it. Details are in the privacy policy.
07 · Waitlist
Get early access
01The pre-order link at the $19 early-access price, before anyone else.
02A note when each drop ships, starting with v0.1 in February 2027.
03Early access to the Groundplane 3D labs as they open.