YakData FDE Project Files

FDE Project Files · YakData

What does a Forward-Deployed AI Engineer actually do?

Starting August 19, follow one real production AI system from business decision through requirements, architecture, build, evaluation, deployment, monitoring, handoff, and improvement.

Project File 001 follows 77 Rules, a live self-serve dashboard-review product used here as the production case. The system combines multimodal AI, explicit rules, deterministic scoring, evidence-backed findings, and user verification or rejection of findings.

Episode 1 publishes Wednesday, August 19, 2026. The full seven-part series will total approximately 5.5 to 6.5 hours when complete.

For technical leaders who need to judge the whole system, experienced builders who want to own more of it, and organizations with a live production problem.

Have a live production problem?Request a Fit Review
7Part production series
~6 hrsWhen complete
1Real production system
5Overlapping role lanes
1 pathDecision to client ownership

Who this is for

Lead it, build it, or bring YakData the live problem.

See the same production case from the perspective of leading the work, building across the system, or solving a live production problem.

Technical leadersKnow what good looks like across the whole system.

Engineering managers · technical PMs · data/AI leaders · architects

The question

“How do I know whether the team is making the right production decisions?”

  • Set a clear decision, definition of done, and ownership.
  • Challenge architecture, integration, evaluation, release, and handoff.
  • Know what evidence should change the plan.
See the technical-leader path →
Organizations with a live problemUse the public work to judge how YakData thinks before you hire.

CTO · CIO/CDO · VP Engineering/AI/Data · operating executives

The question

“Can this problem become a production system our team can run and own?”

  • See the architecture, evaluation, failure, and handoff discipline before a sales call.
  • Review bounded production proof rather than capability claims alone.
  • Start with a $15K Rapid System Definition when the problem needs paid definition.
Work With YakData →
Experienced buildersOwn more than your component.

Senior/staff engineers · AI/ML engineers · data engineers · solutions architects

The question

“How do I move from my part of the system to the production result?”

  • Frame the business problem before broad implementation.
  • Build a thin end-to-end slice across every critical layer.
  • Evaluate, harden, deploy, recover, and hand over the system.
See the builder path →

Not beginner coding instruction. The series assumes working technical experience and focuses on production judgment.

01 · How the work moves

The complete path, not the model call.

Production AI starts with a decision that matters and ends when the system can be tested, run, recovered, and owned.

PM outcome, workflow, scope, acceptance UX human control, review behavior, exceptions FEE application behavior, state, review surfaces BEE APIs, orchestration, rules, security, observability Test/QA acceptance, regression, failures, release evidence

02 · Project File 001

77 Rules: visible production evidence.

77 Rules keeps the series attached to one real system: the decisions, architecture, model boundaries, failures, controls, and handoff stay visible. The case page separates what the live product proves from later client adoption and extension.

77 Rules audit workspace with annotated dashboard review and operating controls
77 Rules audit workspace. Production case used throughout FDE Project Files.

What the case exposes

The model is a component. The system is everything around it.

  • Multimodal perceptionKeep model output structured.
  • Explicit rulesTurn expert judgment into inspectable logic.
  • Evidence disciplineSeparate observation, inference, and missing evidence.
  • Deterministic scoringKeep important scoring under explicit control.
  • Human verificationPut correction and override inside the workflow.
  • Client extensibilityHand over a system the client can run and extend.
What this looks like in practice

A persuasive dashboard may not contain enough evidence to prove the claim. The system sometimes has to ask for context or say “insufficient evidence.”

66% · DIntegrity
87% · BDesign
73% · CCombined
8 / 7 / 35Fail / warning / pass

Public product example: the same Marketing Channel Dashboard looked strong on design while integrity exposed missing comparative context, an unclear time range, and inconsistent monetary units.

Stephen McDaniel, founder of YakData and creator of FDE Project Files
Stephen McDaniel · Founder, YakData

A note from Stephen

I’ll show you the work, including the parts that did not go neatly.

I’m Stephen McDaniel. I’ve spent most of my career in the awkward space between a business problem and the system that eventually has to work. FDE Project Files is where I take that work apart in public: the architecture, decisions, tests, revisions, and tradeoffs that usually disappear from a polished demo.

Sometimes the right answer is more AI. Sometimes it is a rule, a better workflow, a simpler system, or not building the thing at all. I’ll show you the evidence and tell you what I think.

Stephen McDaniel · Founder, YakData

03 · The 7-episode journey

Seven production decisions, from definition to ownership.

Each episode opens on a real artifact, shows the decision at stake and what can go wrong, and closes with the next build, test, release, or handoff decision.

Every episode has to earn its conclusion.Show the artifact.Expose a failure or constraint.Show what changed after evidence.End with the production decision.
Episode 02 · Coming weekly

Turn Expert Judgment Into System Requirements

Turn expertise into rules, evidence, context, workflow states, errors, and testable contracts.

Proof: rule/evidence schemaPreview Episode 2 →
Episode 03 · Coming weekly

Design the Production System and the Work

Design the architecture, frontend/backend contracts, constraints, observability, and overlapping work.

Proof: role/dependency graphPreview Episode 3 →
Episode 04 · Coming weekly

Build the Production-Shaped Vertical Slice

Move one representative case through every critical layer before broadening the backlog.

Proof: vertical-slice architecturePreview Episode 4 →
Episode 05 · Coming weekly

Make the AI Earn the Right to Be Trusted

Benchmark the real task, test critical misses and false alarms, and verify human control.

Proof: benchmark + failure matrixPreview Episode 5 →
Episode 06 · Coming weekly

Turn the Working Slice Into a Production System

Harden validation, retries, versions, regression, telemetry, deployment, and recovery.

Proof: production-readiness checklistPreview Episode 6 →
Episode 07 · Coming weekly

Transfer Ownership and Improve the System

Package the runbook, test evidence, recovery knowledge, extensions, metrics, and next decision.

Proof: ownership/handoff packagePreview Episode 7 →

04 · Project Files

Learn from the artifacts engineers actually use.

Start with the free map and Starter Pack. The $99 Field Library packages the complete 77 Rules case and four specialist working packs. The files are complete; sales are not open yet.

Free · No email

FDE Production Map

The end-to-end production path with role overlap, release checks, and handoff points.

View Production Map
Free · Email

Project File 001

Expanded 14-step production workflow, role matrix, definition-of-done structure, and production checklists.

Get the free Starter Pack
Live production review

Production Reviews

Use the public series and files at your own pace. When a production decision would benefit from structured outside review, bring one sanitized artifact and one focused question.

$1,800 · six monthsFounding price · about 24 live group reviews · opens with 6 members · capped at 12See Production Reviews →

05 · Why YakData

The lessons come from systems I’ve actually built.

FDE Project Files comes from systems I have built through YakData. The detailed proof page shows the work, the numbers, and the boundaries around the claims.

4 weeks77 Rules product build from concept to production. A large non-U.S. bank later adopted it and extended the rule set.
7 weeksCommercial-intelligence system delivered against a 4 to 6 month comparison baseline.
~30 systemsForecasting, analytics, valuation, sponsorship, and audience-intelligence systems in 2.5 years.
$1B+Acquisition decision supported by revenue verification and P50/P90 scenario modeling.

Review the detailed proof record

For teams building this now

This is the work.

If your team is trying to turn an AI experiment into something people can rely on and own, YakData handles the definition, build, evaluation, deployment, and handoff.