Getting started, questions answered.

An adaptive AI product is one where the interface itself — not just the recommendations inside it — reshapes to who the user is, what they are doing, and what they need next. The thesis is that software should learn the human, not the other way around.

No. Personalisation typically changes content inside a fixed layout (which products you see in a row, which articles in a feed). Adaptive AI changes the layout, the depth, the copy, the primary actions, and sometimes even the pricing — not just the contents of a carousel. It is a categorical jump in the surface area of what adapts.

We are model-agnostic and choose based on the workload. Reasoning is typically done on frontier models such as Claude Opus or Sonnet, or GPT-class equivalents. Fast classification and routing run on smaller models. We are explicit with clients that model choice is rarely the moat — the event taxonomy, the typed render schema, and the reward loop are.

Most first pilots are six to twelve weeks, depending on industry and the complexity of the surface being adapted. Consumer-facing surfaces (onboarding, search, discovery) tend to be at the shorter end. Regulated or operational surfaces (clinician tools, control-room software) tend to be at the longer end.

Pilots typically run from a few thousand dollars for tightly-scoped work to mid double-digit thousand dollars figures for larger pilots that include reference architecture, evaluation harnesses, and team enablement. We are transparent about pricing in the very first conversation; no surprises later.

Almost always, yes. The most successful engagements have us embedded with the client team rather than working in isolation. We design the typed contracts, the inference layer, and the first adaptive surfaces, while the client team owns the integration and long-term operation.

You do. Our standard engagement terms assign all delivered work product and the surrounding IP to the client. We retain the right to talk about the engagement in anonymised form, and the right to reuse generic patterns and architectures.

We treat client and end-user data with care that meets or exceeds the standards of regulated industries. We never train shared models on client data. Where required, we deploy entirely within the client’s cloud environment. Our full security posture is on the Security page.

Adaptive systems carry the same risks as any AI system, and a few that are specific to them — specification gaming on engagement metrics, for example. We take this seriously: every engagement includes an explicit safety and fairness evaluation, with offline replay rigs and continuous monitoring once deployed.

Yes. The home page is calibrated to provide useful content even at the curious or academic depth. If you are doing research in this area, we are happy to chat — we publish occasionally and have hosted research visits in the past.

Our reference architectures, evaluation harnesses, the typed-render component library, and our industry-specific playbooks are reserved for paying clients and active conversations. Everything you need to understand the field, evaluate it for your context, and decide whether to engage us, is free.