Scope & Disclaimers

What Ground Floor is

Ground Floor documents experiments with open-weight large language models running on local Apple Silicon hardware. It's an education project, not a product, not a consulting firm. The site covers: which models run on which hardware, at what speed, with what output quality, for specific tasks in regulated industries.

Every experiment is a technical test. The verdict tells you whether the technology is capable of a specific task at a specific level of quality. Nothing more.

What Ground Floor is not

Not legal advice. Nothing on this site constitutes legal advice, and no attorney-client relationship is formed by reading or interacting with this content. If you are an attorney considering local LLM tools for your practice, consult your state bar's ethics counsel before deployment.

No compliance advice. Ground Floor tells you what the technology can do. Whether it should be part of your workflow, given your jurisdiction, your specialty, your client agreements, your licensing requirements, is a different question. One with many variables this site isn't equipped to answer.

Not medical advice. Nothing here is clinical guidance. Experiments involving medical documentation are technical tests of AI output quality, not clinical recommendations. Any AI-assisted documentation in a clinical setting requires review and accountability by a licensed clinician.

No financial advice. Nothing here constitutes investment advice, financial planning guidance, or recommendations about products or services for your clients.

The key distinction

What Ground Floor does is document the technical foundation. Your specific compliance situation, your state, your county, your specialty, your licensing body, your malpractice carrier, your contracts, is a conversation with someone who knows your jurisdiction. That person is not this website.

To make this concrete: an experiment showing that a local LLM can draft SOAP notes at sufficient quality for clinician review is a technical finding. Whether using that setup in your specific practice satisfies HIPAA, your state's medical privacy laws, your EHR vendor's terms of service, your malpractice insurer's requirements, and any applicable CMS guidelines is a separate, jurisdiction-specific, practice-specific question. The technical finding is useful context for that conversation. It is not a substitute for it.

Air-gapped does not mean compliant

A common misread of air-gapped AI: that running a model on hardware you own automatically satisfies data privacy requirements. This is not correct.

Running air-gapped eliminates a specific category of risk: third-party processor exposure. It simplifies the compliance picture significantly. But it does not replace a written information security program, staff training, proper access controls, business associate agreements with other vendors you use, incident response plans, or any other element of a regulatory compliance program.

Air-gapped is a necessary condition for certain compliance postures. It is not a sufficient condition for any of them.

Accuracy and currency of experiment results

Every experiment was conducted at a specific point in time with a specific hardware configuration and specific model version. Results may vary based on: model version updates, hardware differences, operating system and runtime differences, input quality and format, and dozens of other variables.

Experiment verdicts reflect conditions at the time of publication. The date of each experiment is shown prominently. Before relying on a result, check whether newer models or hardware are available that change the picture.

No endorsements

Specific products, models, and vendors are mentioned because they're what I actually use. Not endorsements. Ground Floor has no commercial relationships with any hardware manufacturer, model developer, or software vendor mentioned here.

Who I am in relation to this work

I am a fellow experimenter, not a credentialed professional in any of the industries this site covers. There is no law license, no medical license, no financial advisory registration, or any other professional certification behind me relevant to the regulated industries documented here.

What I have is technical fluency: I understand how these models work, what hardware configurations enable specific use cases, and how to run controlled experiments that produce reproducible results. That is the scope of my contribution.

You experiment at your own risk. Every result on this site is a technical finding from a private lab. Before applying any finding to a real practice workflow, clinical, legal, financial, or otherwise, you are responsible for evaluating whether it is appropriate for your specific situation, your jurisdiction, and your professional obligations. I am not a substitute for that judgment. Treat me as a starting point for it.

Contacting me

I'm happy to discuss the technical side of air-gapped AI deployment: how specific models work, what hardware configurations enable specific use cases, and the current state of open-weight tooling for regulated industries.

I cannot evaluate your specific compliance situation, review your existing procedures, or tell you whether a given setup satisfies your regulatory obligations. That's not a hedge. It's a genuine limitation. Compliance in regulated industries is jurisdiction-specific, practice-specific, and changes over time. It requires a professional who knows your situation.

Reach me on LinkedIn.