House Rules

The Operator’s Framework for AI Deployment

Ten years running restaurant floors, a few years running my own online retail business, two years shipping AI apps. It has been the same job every time. Keep the system up while people are counting on it.

Run the diagnostic See the builds

01The zigzag

This does not look like a normal resume

Here is the order it happened in, and what each job taught me.

Bartender at a pool hall, then the de facto manager with the keys whenever the owner was away.
Helped open a new location for a national grill chain, then supervisor with opening keys.
Manager at a busier location in the same company. Bar first, then service, then marketing, then the kitchen.
Bar manager to service manager to assistant general manager at two higher volume restaurants. I left the industry when COVID hit.
Ran an online retail business for a few years, then moved to building AI products. 12+ live products.

02The incident

The Sunday lunch rush

Sunday lunch is one of the two or three busiest shifts of the week. That Sunday the point of sale went down. No tickets, no card payments, nothing.

I stopped seating people for a minute and made one decision: every guest hears it from me before they sit down, not from a server after they have eaten. We took cash where people had it. We handed out gift cards where they did not. I kept the kitchen cooking off handwritten tickets and I touched every table in the dining room, twice if the food was slow.

Comps ran 20 to 30 percent of the day’s sales. That is an ugly number and I would take it again. Nobody left angry.

The same shift, translated

What a restaurant outage and a stalled AI rollout have in common, and the move in each case.
Entry	On the floor	In an AI rollout	The move
01	The point of sale goes down mid rush	A model or an API degrades in production	Tell the people it affects before they find out on their own
02	The kitchen keeps cooking, the front cannot ring anything in	One subsystem is fine and the handoff between them fails	Go find the broken handoff, not the broken team
03	Comps reach 30 percent of sales	Cost or latency spikes	Decide what you will eat today to keep the trust
04	Every table gets touched	Every stakeholder gets a status	Keep talking until the system is back
05	After the shift: what do we change	After the incident: what do we instrument	Write it down while it still hurts

On the floor
The point of sale goes down mid rush
In an AI rollout
A model or an API degrades in production
The move
Tell the people it affects before they find out on their own
On the floor
The kitchen keeps cooking, the front cannot ring anything in
In an AI rollout
One subsystem is fine and the handoff between them fails
The move
Go find the broken handoff, not the broken team
On the floor
Comps reach 30 percent of sales
In an AI rollout
Cost or latency spikes
The move
Decide what you will eat today to keep the trust
On the floor
Every table gets touched
In an AI rollout
Every stakeholder gets a status
The move
Keep talking until the system is back
On the floor
After the shift: what do we change
In an AI rollout
After the incident: what do we instrument
The move
Write it down while it still hurts

03The diagnostic

The Last Mile Diagnostic

For a rollout stuck between “we bought it” and “people use it and it pays off.”

Seven questions. It names the bottleneck that is most likely holding you up and tells you what I would do about it first. It runs in your browser. Nothing is sent anywhere and nothing is stored.

1 of 7

Who owns this rollout day to day?

Keys 1 to 4 select. Enter advances. Backspace goes back.

04Proof of work

Two live products, and what each one cost me

No. 01

Ripple News

Product: A news app that rewrites each story for the reader's profile.
Lessons: Three platform rewrites: web, then React Native, then native Swift. I rebuilt the image pipeline after it shipped distorted and missing images, and I rebuilt the content pipeline on compliant sources. Then I cut that pipeline from every 30 minutes to every 8 hours once it was clear this is not a breaking news product, and the run cost came down with it.

App Store Read the case study

No. 02

Yapper

Product: An AI answering service for small trades businesses. It catches the missed call, collects the caller's details, and texts the owner a summary so the job does not disappear.
Lessons: The build was the fast part. Distribution has taken longer: cold email, search ads, and postcards to plumbing and HVAC shops. Nobody in that trade is shopping for AI. They are counting the jobs they missed last week.

tryyapper.com Read the case study

The rest of what I have shipped

05Fit

Where this lands

Role 01

Technical Deployment Lead

“CEO of your engagement” is the Sunday I ran with no point of sale. I decided what we ate, who got told, and in what order, and I owned the number at the end of it.
“Create structure where none exists” is Ripple. Three platform rewrites, an image pipeline rebuilt after it shipped broken, and a content schedule cut from 30 minutes to 8 hours once the evidence said so.
“Define success metrics and measure ROI” is the diagnostic above. It exists because most stalled rollouts I read about never wrote down what done means.

Role 02

Enterprise Account Executive, Retail

“Store operations” is ten years of it. Labor, inventory, vendors, a P&L, and the shift that goes wrong at the worst possible hour.
“Consultative” is what the diagnostic does. Seven questions, no pitch, and a specific first move at the end.
“Retail pain points” is who I sell to now. Yapper's customers are one or two person businesses who lose money every time the phone rings and nobody picks up.

“We urge you not to exclude yourself prematurely.”

I am not.

Email me Download resume