Case Study: Installing a Private AI Assistant for a Business Owner in South Florida
AI & Productivity

Case Study: Installing a Private AI Assistant for a Business Owner in South Florida

Ricardo Gil
November 19, 2026
6 min read
#Private AI #Local AI #Case Study #Mac mini #AI Agents #South Florida #Self-Hosting

A business owner in South Florida asked ChatGPT who could install a private AI assistant in his home office. ChatGPT pointed him to this site. A few weeks later the system was running on a Mac mini on his desk and had passed every test against his real documents. This is a write-up of that project. The client's name, location and business details are kept private.

At a glance

  • Client: owner of several companies in South Florida
  • Goal: a private AI assistant with one agent per company and one orchestrator agent he talks to
  • Hardware: an Apple Mac mini supplied by the client, plus a Raspberry Pi
  • Team: me on site, and the client's own technical lead working remotely
  • Timeline: a discovery call, then two on-site visits in the same week
  • Result: every acceptance test passed on site, followed by 30 days of support

The problem

He runs several businesses at once. He already used ChatGPT every day, and a lot of his work happens in WhatsApp groups. What he didn't want was to keep pasting company documents into a cloud chatbot. He wanted an assistant that knew each business, ran on hardware he owned, and was still easy to use from his phone.

The design he had in mind: one AI agent per company, each with its own documents and context, and an orchestrator agent on top that he talks to and that sends each request to the right company's agent.

How we split the work

The client already had a technical lead who designed the agent architecture and works with him remotely. After the discovery call I prepared two quotes, one built on his technical lead's plan and one with my own recommended approach. We agreed to run the project in phases.

Phase 1 was my part: the on-site work. That meant installing and configuring the hardware and local components, connecting the pieces the remote lead needed, and testing everything with the client in the room. Keeping Phase 1 self-contained meant it could be finished, tested and signed off on its own.

What we installed

A Mac mini as the private AI server

The Mac mini is the core of the setup. It runs the local AI components and stores the documents the agents work from, on hardware the client owns. It's quiet, small enough to sit on a desk and light on power, which matters for a machine that stays on around the clock.

A Raspberry Pi for remote unlock

An encrypted Mac doesn't fully come back on its own after a power outage. It waits at the unlock screen until someone enters the password. In South Florida, where storms regularly knock out power, that's a real failure mode for a home server. We added a Raspberry Pi on the local network so the Mac mini can be unlocked remotely after a restart, without anyone having to be in the house.

A secure bridge to the tools he already uses

He wasn't going to switch to a new app. Together with his technical lead we set up a secure tunnel so the system can be reached from ChatGPT, the interface he already uses every day, instead of asking him to learn something new.

Access from his phone

We finished with a live demo on his iPhone, so he left with an assistant he could use from anywhere, not just a box on a desk.

Testing with real documents

Demo files make every AI system look good. We tested with real documents from his businesses, on site, with him watching, and went through the acceptance checks one by one. Everything passed on the second visit. The handoff included final documentation and a 30-day support window.

What I'd tell other business owners

  • Start from the tools people already use. If the owner lives in ChatGPT and WhatsApp, the system should meet him there. Getting him to use it matters more than the architecture.
  • Plan for power outages. A private AI server is only useful if it comes back on its own, or can be brought back remotely.
  • Test with real data before calling it done. Sample files hide the problems you'll actually hit.
  • Split the project into phases that can each be tested and signed off.

Who this setup fits

Owners and teams who handle information they don't want sitting in a public chatbot: law firms, medical and dental practices, accountants, family offices, and owners running several companies. I install on site across Miami-Dade, Broward and Palm Beach counties, work remotely anywhere else, and work in English or Spanish.

Frequently asked questions

Can I run a ChatGPT-style assistant privately in my office?

Yes. A private AI stack runs language models on hardware you own, so your documents live on your machine instead of a third-party service. You can still connect it to tools you already use, as this client did with ChatGPT. My guide to running Ollama and Open WebUI on a mini PC shows the basic building blocks.

What hardware do I need?

For one person or a small team, a Mac mini or a capable mini PC is usually enough. Bigger teams and bigger models need more memory or a GPU. I size the hardware on the discovery call. For options, see my mini PC guide for local AI.

How long does an installation take?

This one took a discovery call and two on-site visits in one week. Timing depends on how many systems need to be connected and how many people will use it.

How much does it cost?

My Private AI Stack Deployment starts at $2,500 as a one-time setup, with optional monthly support.

Do you work in Spanish?

Yes. I'm a native Spanish speaker and handle installs, training and documentation in either language. Read this case study in Spanish.

Want a private AI assistant for your business?

I install private AI systems on hardware you own, on site in South Florida or remotely, documented and handed off. See the Private AI Stack →

Not sure what you need? Book a free discovery call.

Ricardo Gil is a full-stack engineer (.NET/Angular) with 6+ years of experience, based in South Florida. He installs private AI systems and automations for businesses, and runs local LLMs on his own Proxmox homelab. More about Ricardo →
📬Weekly Newsletter

Get the best home lab & AI content

No spam. One email per week. Unsubscribe anytime.

Share this article