xcelMeGary Bennett Let's talk
Mobile & AI development · Scottsdale, Arizona

Apps people use every day. AI that keeps your data yours.

I'm Gary Bennett. I've built Apple apps for brands like Life Time and Choice Hotels, co-written Apress's beginner books on Objective-C and Swift, and now help teams add AI to their products, including private models that run on hardware they control.

Send a short note and I'll set up a free 30-minute call.

mac-studio — llama-server
$ llama-server --model nemotron-3-super \
    --ctx-size 65536 --port 8080

model    Nemotron 3 Super 120B-A12B
quant    UD-Q4_K_XL · 84 GB
machine  Mac Studio · M5 Max · 128 GB
access   private tunnel, no open ports
routing  local model or Claude, per question

✓ listening · your data stays here 
Apress author · Objective-C & Swift booksApple developer sample codeU.S. Navy nuclear engineer, 10 years
What I can do for you

Four ways to work together

Bring an app idea, an app that needs rescuing, or a pile of data you want to ask questions of.

Mobile apps

Build or rescue a native Apple app

An iPhone, iPad or Mac app that passes review and that people keep using. New builds, or taking over an app another team left behind.

  • Swift and SwiftUI, built to Apple's guidelines
  • Location, Bluetooth and hardware integration
  • TestFlight, App Store submission and release automation

Good fit: a product team or a business with an app idea or an aging app.

AI features

Add AI to your product

Assistants and features that answer from your own data, with read-only access, budgets and audit logs designed in from day one.

  • Chat over your logs, documents, tickets or code
  • On-device AI in your iPhone app with Apple's models
  • Claude-powered agents for your internal workflows

Good fit: teams with useful data that's hard to get answers from.

Training

Hands-on AI workshops

Practical sessions built around your team's real tools and data, from someone who has taught technology courses since 2009.

  • Prompting and AI tools for everyday work
  • AI coding agents for developer teams
  • Custom curriculum for your industry

Good fit: teams that have AI licenses but aren't getting much from them.

Why private AI

Use the right model for each job

The best AI setup for most businesses isn't one model. It's a choice made per question: the strongest cloud model when it matters, your own hardware when the data is sensitive or the volume is high.

Your data stays yours

Patient records, contracts and logs never leave your network.

Costs you can predict

Pay once for hardware instead of a bill that grows with every question.

No lock-in

Open-weight models and standard APIs, so you can switch models as better ones arrive.

Hardest reasoningClaudeMulti-step analysis, code questions, long reports.
Private or high volumeYour local modelSensitive data and bulk jobs with no per-token bill.
In your appApple on-deviceMLX and Foundation Models for features that work offline.
Results

Recent work

Healthcare · iOS

Hospital vendor badge app

TrueAccess · Swift · CoreBluetooth

Challenge

Vendors visiting hospitals need to check in at the right building, even when several hospitals share one address.

What I built

Geofencing with polygon detection for each building, and a wearable Bluetooth LED badge that shows access status at a glance.

Outcome

An app that now ships under partner brands, with a logging pipeline that turns field reports into fixes.

AI · Support

Log analytics assistant

Claude · local LLM · Postgres

Challenge

Answering a support question meant an engineer digging through app logs and the source code.

What I built

A chat assistant over the logs and a code mirror, with a read-only database role, one query at a time, budgets and an audit log.

Outcome

The support team asks in plain English. Each question can run on Claude or on the private model.

Productivity · on-device AI

MyRunBook Planner

SwiftUI · Vision · MLX

Challenge

Tasks, meeting notes and receipts were spread across too many apps.

What I built

One app for iPhone, Mac and Vision Pro that reads receipts with on-device OCR and pulls tasks out of meetings with a local model.

Outcome

Meeting transcripts never leave the Mac, and nothing gets lost between apps.

More work →

How it works

From first call to shipped

Talk it through

A free 30-minute call about what you need and whether I'm the right fit.

Plan and prototype

A written plan and a working prototype you can tap through before the full build.

Build and ship

Weekly builds on TestFlight or a staging site, then release.

Support and hand-off

Monitoring, fixes and documentation so your team can own it.

Questions

Common questions

Can AI features work without sending our data to the cloud?

Yes. Small models run on the iPhone itself with Apple's on-device frameworks, and larger ones run on a server in your office. I set up the private model, connect your apps to it, and route only the questions you choose to a cloud model like Claude.

What hardware does a private model need?

It depends on the model and how many people use it. A Mac Studio with 128 GB of memory runs a 120-billion-parameter model comfortably for a small team, and smaller models run on far less. I'll size it to your use before you buy anything.

Will you take over an app another developer built?

Yes. Much of my work is improving existing Swift apps: finding the bugs field reports point to, modernizing old code and getting releases moving again.

Do you work with small companies?

Yes. I've worked with national brands and with small businesses. The process scales down: a short plan, a prototype and regular builds.

Can you train our team instead of building for us?

Yes. Workshops can stand alone or come with a project, so your team keeps building after I hand it off.

Have an app or an AI idea?

Send me a short note on LinkedIn about what you're trying to do. We'll set up a free 30-minute call, and you'll come away knowing whether it's a fit, roughly what it takes, and what to do first.