|
Download README.md from AIR4T/README: direct link, hf CLI and curl.
- Browser
- Download file 3.12 kB
-
https://huggingface.co/spaces/AIR4T/README/resolve/main/README.md
- Command line
-
hf download hf://spaces/AIR4T/README/README.md
-
curl -L -o README.md https://huggingface.co/spaces/AIR4T/README/resolve/main/README.md
3.12 kB
| # AI Revolution 4 Technologies Pty Ltd | |
| **Sovereign AI for regulated Australian work.** | |
| Sydney, Australia · ABN 27 680 691 589 · [air4t.ai](https://air4t.ai) | |
| --- | |
| ## The position | |
| In regulated work the binding constraint is not model capability. It is **verification cost**. | |
| A model that is right 95% of the time is not usable by a solicitor or a specialist physician if | |
| checking the output costs more than doing the work. And when a court or a regulator allocates the | |
| duty of verification to the practitioner personally, verification stops being a product feature | |
| and becomes the product. | |
| So we build systems where **the registered professional decides and the AI prepares** — drafting, | |
| retrieving, triaging and organising — with the reasoning visible and the evidence auditable. Not | |
| because autonomy is technically out of reach, but because in these domains an unverifiable answer | |
| has negative value. | |
| ## What we work on | |
| **Statutory grounding.** Retrieval and citation attestation over Australian primary law, so an | |
| assertion about a statute can be traced to the provision, at a point in time, with the currency of | |
| that provision established rather than assumed. | |
| **Privilege-preserving architecture.** Legal professional privilege in Australia can be lost by a | |
| disclosure that is unintended and transient, and the loss falls on the client rather than the | |
| firm. We treat that as a structural question, not a risk-weighted one — the boundary is enforced | |
| in code, not in policy. | |
| **Data residency and local-first inference.** Australian data stays in Australia where the law or | |
| the client requires it. Where no compliant hosted path exists, the answer is local inference, not | |
| an exception. | |
| **Governed agentic systems.** Deterministic controls around probabilistic components: human | |
| approval gates on consequential actions, tamper-evident audit, and confidence treated as a safety | |
| signal rather than a marketing number. | |
| **Domain embeddings and continual learning** for Australian legal, medical-administrative and | |
| security corpora. | |
| ## How we work | |
| - **Code decides, models advise.** Anything requiring certainty — governance, security, | |
| compliance — runs on deterministic code. The model advises; it does not adjudicate. | |
| - **Honest amber over false green.** A check that cannot run reports *unknown*, never *pass*. An | |
| unverifiable claim is a blind spot, and reporting it as success is how systems come to lie. | |
| - **A saving is claimed only when measured.** Numbers in our documentation are measured or absent. | |
| - **No training on customer data.** Client and patient material never trains a model — ours or | |
| anyone's. | |
| ## Status | |
| This organisation is new. **No public models or datasets yet.** Anything published here will be | |
| released deliberately, with licensing and attribution settled first — particularly for Australian | |
| legal corpora, where the licence position differs by source and by field. | |
| We would rather have an empty profile than a published artefact we had not cleared. | |
| --- | |
| *Interested in sovereign AI for Australian legal, medical or defence work? | |
| [air4t.ai](https://air4t.ai)* | |