Designing for a non deterministic product

Research tools live or die on trust. When the output is generated by an AI, trust has to be designed in deliberately. The platform needed to surface complex, source backed research fast, while giving users the confidence to act on what they were seeing. That tension between speed and verifiability became the central design challenge.

Business Goals

Deliver an AI powered research platform that reduces time on task for knowledge workers, supports tiered access across user types, and positions the product for a phased commercial rollout.

Strategic Input

Played a key role in feature prioritisation, balancing client must haves against delivery timelines across a staggered release plan. The decisions we deferred were as important as the ones we shipped.

Deep discovery, short sprints

The process started with an extensive UX discovery phase before wireframes were touched. Stakeholder interviews, a value proposition canvas, competitor and cross market research, lightning demos, How Might We statements, and structured ideation sessions. That groundwork made the design sprints sharper and faster.

My Role

Part of a two designer team embedded across the full product lifecycle. I led the wireframing and UX structure for key workflows, presenting design rationale directly to senior stakeholders at twice weekly feedback sessions. I also took ownership of the design system, building a hybrid on top of Untitled UI — extending and adapting its patterns to meet the specific demands of an LLM powered product the existing system wasn't built to handle.

Design System

The client had an existing system but it wasn't fit for purpose. Rather than start from scratch, I used Untitled UI as a foundation and built a hybrid — retaining what worked, extending what didn't, and layering in design tokens and a semantic colour architecture designed for theming from day one. A pragmatic call that saved time without compromising on where it mattered.