Skip to content

How the pilot works

A focused deployment, built and evaluated with the people who will use it.

From discovery to everyday use

Each stage has an agreed outcome, from selecting the work to handing it over.

  1. 01 / Understand

    Choose the work.

    Agree the workflows, sources and starting measures.

  2. 02 / Implement

    Build the system.

    Configure hardware, connect information and test the agents.

  3. 03 / Use

    Train the team.

    Work through real requests and gather feedback.

  4. 04 / Evaluate

    Put it in your hands.

    Review the results and hand over the agreed workflows.

Before we begin.

Will the hardware definitely be a DGX Spark and Mac mini?

That is an example configuration. We may agree a different, comparable setup when it better suits your workloads. The final specification is confirmed before you commit.

Explore the private compute
Does all our information stay local?

A local-only setup keeps model processing on your infrastructure. If you choose a hybrid workflow, approved requests are sent to an external model provider under the agreed data and routing rules. That choice is explicit.

See local and hybrid architecture ↗
How do you size the hardware for the team?

Around five people is the intended pilot size. We validate the chosen models, document volumes and expected simultaneous use before agreeing the setup.

Do we need to know which agents we want?

No. Start with the recurring work, the information it uses and the people involved. The audit helps select tasks that are practical to implement and useful to evaluate.

How long does the pilot take?

Three months is an illustrative example. The actual schedule depends on workflow complexity, access to your systems and your team’s availability. The milestones and licence start date are agreed before kickoff.