PerennialA11y - Digital Accessibility & Responsible AI Engineering

Responsible AI Engineering

AI your nonprofit runs on its own computers.

I help ethically conscious nonprofits use local AI instead of cloud AI. Local means the model runs on a computer your organization owns, and your documents stay on that computer. I set it up for office work, for development work, and for the people on your team who are not technical and want to use AI day to day.

By responsible I mean things you can check. Your files stay on a computer you own. You can leave without losing your work. A person decides, not the model. I test it on your real tasks before you rely on it, and I tell you when cloud AI is the better fit.

I'm starting with nonprofits and will work with other kinds of organizations later.

Two things first, because you should hear them before the pitch. Google gives eligible nonprofits Gemini at no cost, and local AI will not beat free. And the strongest cloud models are better than any model I can run on a computer in your office. If you choose local AI, it is for what you value: where your files live, who can read them, and whether you can leave a vendor without losing your work. If free cloud AI fits your values, use it. I'll tell you so.

Nonprofits that want to keep their data, their model, and their exit. Usually that means one of two things. Your work touches files you would not paste into a cloud chat window: case files, client notes, health or immigration records. Or your organization would choose local AI on principle, even for ordinary office work.

Not a nonprofit, and want the same thing? Email me anyway.

Local AI plan

About one week, fixed price. I spend 20 minutes with each person who would use it and ask what part of their job they would happily never do again. I map where your data goes today and what could run locally, and I time the tasks people repeat. If you already own a computer that could run a model, I test one on it. You get a build or don't-build recommendation, walked through with you live. "Don't build" is a real answer.

Office work

A local model on a computer your organization owns, set up in about a week. Your staff get a shared chat window in the browser for drafting, rewriting, and summarizing, and for asking questions about documents you give it, with the source cited in the answer. Staff reach it from their laptops and phones over an encrypted network that only the devices you allow can join. Before anyone relies on it, I test it on 10 to 20 of your real tasks and write down what passed, what failed, and which tasks should stay where they are.

Development work

For the developers or volunteers who build your website and tools: a coding model that runs on your own computer, so your code goes to a machine you own instead of a cloud service. I can also build local models into your own software. Local coding models are weaker than the top cloud ones. I'll show you where, on your own code, before you decide.

Day-to-day help for staff

A tool nobody uses does nothing. I train your staff for half a day on their own work, not on demo prompts, and everyone gets a one-page guide to what it can and cannot do. After that, a monthly care plan covers model updates, re-testing on your tasks, new uses as they come up, and questions from staff. You can cancel it any month. The written runbook lets you run without me.

Every engagement is quoted as a fixed price before work starts. Your organization buys the computer directly, or I pass it through at cost. When I leave, you keep the computer, the model, and the runbook. No lock-in, not even to me.

It can draft, rewrite, and summarize. It can answer questions about documents you give it, with sources. It can help with code.

It cannot make decisions on its own. It drafts and answers, and a person decides. It does not match the top cloud models on every task. I tell you which tasks, in writing, before you buy.

There are things I won't build: anything that watches your staff, a copy of a real person's voice or face without their consent, or anything that sends your data to a third party without telling you.

My own local AI setup

Reference build I run every day, 2026

In 2026 I moved my own AI work off cloud services and onto computers I own. Web search, chat, and coding tools run on local models on my machines, and my family uses them too. I sell what I operate.

  • Local models on my own Apple silicon server, reachable only over an encrypted network that my own devices can join, with the disk encrypted.
  • An evaluation harness built before I trusted any of it. A local model takes over a job only after it proves, on my real work, that it can do that job.
  • Local-first routing, with written rules for what may ever fall back to a cloud model, so work slows down instead of breaking.
  • Fixed the failures that stop real setups: power loss, reboots, start on boot, and runtime updates. I found these by hitting them.

This is my own setup. I have not done a nonprofit install yet. The first one will be priced and run as a first.

Mir'at

Personal AI, in development

The personal AI I'm building for myself first: pages about my life that I own and can check. It is where I test the ideas I bring to client work. In development. No public beta yet. Mir'at site, mirat.studio

I'll tell you when AI is the wrong answer.

Not every organization needs local AI, and not every problem needs AI at all. My plans recommend the least powerful tool that serves you, including "none," and when cloud AI under a proper agreement is honestly your best fit, I'll say that too. The evaluation comes before the build, and the recommendation is not tied to what I'd like to sell you. AI should serve us and humanity not the other way around.

If AI is part of how you think, it should belong to you.

I'm neurodivergent. AI tools are part of my working memory, my planning, and my communication, which means they hold things I would never hand to a vendor. The same holds for an organization. Your clients gave their files to you, not to an AI vendor.

I'm a developer with over a decade in production.

Local AI is not just a model on a box. It's epistemology, reliability, security, privacy, disaster recovery, interfaces, performance, accessibility, and UX you'll actually want to live in. I've shipped production software for over ten years across startups (including my own), Fortune 500s, and a non-profit, and I bring all of that to these builds, not just the AI parts.

Everything I ship is built to be accessible.

I'm also an accessibility engineer, so every interface I deliver is built with WCAG standards as the floor and real accessibility as the goal from day one: usable by your staff and the people you serve, including people with disabilities. See my digital accessibility engineering services.

Interested in running AI on your own computers?

If your nonprofit holds files that shouldn't leave your hands (your clients', your staff's, your donors') and you'd rather run AI on computers you own, I can help. Email me to book a free 30-minute call. I'll listen first. If local AI isn't the right fit, I'll tell you that for free.

Email me