Aprentiz home Apply for a founding research place
Menu

Phil O’Brien

Founder, Aprentiz

Building the evidence infrastructure that regulated firms need before they can trust AI with consequential work.

Phil O’Brien, Founder of Aprentiz

AI will reshape regulated work. Evidence will decide whether it deserves trust.

“Build the evidence first. Make the claim only when it resolves.”

Phil has spent twenty-five years building and running businesses. He founded Aprentiz around a practical observation: in high-stakes work, an answer is only useful when the firm can understand what applies, inspect the evidence and account for the decision that follows.

That conviction drives a substantial regulatory-intelligence stack: source ingestion, private inference, hybrid retrieval, governed memory, evidence design and operating controls.

As project lead for Aprentiz’s UKRI AIRR work, Phil has directed controlled retrieval and domain-adaptation research on Isambard-AI, part of the UK’s AI Research Resource.

How Aprentiz is led.

A serious engineering culture for consequential systems.

  • Evidence before assertion

    The system must preserve what is known, what is missing and what needs human judgement before it makes a consequential claim.

  • Engineering before theatre

    The work starts with identity, data boundaries, provenance, failure states and recovery, not a polished answer box.

  • Ambition with discipline

    We pursue difficult RegTech problems with controlled research, explicit gates and a refusal to overstate the result.

One lab, one accountable system.

We join regulatory knowledge, a firm’s own way of working and engineering controls, so regulated teams can make progress without losing the thread of accountability.

  • Regulatory intelligence

    Turning source material, applicability and evidence into a reviewable decision path.

  • Private AI systems

    An architecture that runs inside controlled data boundaries, with no fallback to an outside AI service.

  • Research-grade engineering

    Controlled evaluation and national-scale compute, used to improve the system without confusing research with product proof.