First pilots / qualification open
Bring a model, a workflow or a research question, and the result you need to achieve.
We’re taking enquiries for our first service projects, joint tool development and research trials. Before work starts, we agree what success looks like, what needs changing and when to stop.
Three different commitments
Choose a service for an immediate technical decision, co-development for a tool your team needs, or research for a question that requires an experiment.
Start with the 10-Day AI Fit Test: A$5,280 including GST (A$4,800 ex GST). Can technicians search approved maintenance records on the existing workstation? Test one useful workflow against real operating limits.
Help test a deployment, training or evidence-review tool in your environment, with the people who would operate it.
Test a model conversion, selective execution method or native ERAIS profile against an agreed baseline. Measure the quality and resource use on the task that matters to your team.
A good first pilot
A good first pilot is small enough to control and important enough to change a decision—even when the answer is no.
The decision owner, lawful access, representative non-sensitive cases, target environment, and people who can witness the result.
Rights, data handling, funding, roles, baseline, acceptance criteria, security, publication, and stop conditions close in writing.
The owner sees the agreed cases, evidence, and failures—then chooses deploy, change, remediate, rescope, or stop.
Commercial pilot options
The 10-Day AI Fit Test is one starting point for a workflow trial. Model conversion and research questions can be scoped directly around the relevant comparison.
What useful AI can run inside your actual data, hardware, connectivity, power, and accountability limits—and is deployment worth funding?
For A$4,800 ex GST, inspect the real system boundary, run a small local demonstration, and leave with the evidence to deploy, change the model or hardware, remediate—or stop.
Audit and local demonstration only. No production deployment, full-corpus ingestion, bespoke integration, continual learning, certification, penetration test, production SLA, or guarantee of accuracy, savings, deployability, or ERAIS performance.
Can one valuable AI workflow run usefully inside your approved local hardware and data boundary?
Prove—or disprove—that one useful workflow can run locally, then leave with the tested system and the evidence behind the decision.
No production high availability, public exposure, autonomous action, continuous learning, broad fine-tuning, certification, 24/7 support, or unbounded integration.
Can one AI release be reconstructed and defended technically?
Reconstruct one AI release end to end—what it claims, what it used, what failed, what changed, and why the decision should be release, remediate, or hold.
Technical evidence and release reconstruction only. No legal opinion or privacy, clinical, safety, cyber, engineering, or regulatory certification.
What does one controlled experiment say about a material technical or commercial decision?
Turn one important uncertainty into a pre-agreed experiment that a decision owner can act on and a reviewer can inspect.
One question, one controlled protocol, one decision. No open-ended R&D programme, guaranteed positive result, or automatic product-performance claim.
If the Fit Test passes
The results identify the changes, operating costs and support needed for the next stage.
Map the limits and show what works on the target machine.
Deliver one workflow using the best-fit qualified model and runtime.
Train the operator. Test every change. Keep restore and rollback available.
Earn repeat sites, support, reusable software, or licensing from real use.
Product co-development
Explore these three tools in development. Each page explains the working prototype and the tests required before delivery.
Could your team run a local model behind a controlled endpoint?
Can one approved workload use a controlled local endpoint with the required operational controls?
Test one approved local-model workload behind a controlled endpoint with explicit identity, capacity, audit, restore, and rollback behavior.
One approved workload and supported configuration. No OpenAI-wide compatibility, general availability, production SLA, or customer-operability claim before the gate passes.
Can someone else repeat the training and check the result?
Can a second operator reproduce one bounded data-to-evidence workflow?
Make one local AI training workflow reproducible from approved data to a versioned checkpoint, evaluation, and evidence pack—then prove a second operator can run it.
One registered data-to-evidence workflow, environment, operator cohort, model/profile, rights set, and acceptance protocol. Not a general training platform.
Can your next reviewer follow the work?
Can a repeated release-evidence process become a reviewer-operable workflow?
Turn one repeated AI release-evidence process into a controlled path from claim to artifact to test to approval—then prove a second reviewer can follow it.
One repeated release-evidence process, one minimum schema, file or bounded API-export integration, and one second-reviewer workflow. Not a generic governance or accreditation platform.
Try it on your work
Tell us what you’re trying to do and where you get stuck. We’ll work out whether a small, practical test makes sense.