Proxima — the first AI-native platform in proptech.
Director-led redesign built when no interaction playbook for this domain existed. Conversational interface behaviours, AI UX patterns, and a full design system from scratch, across 57 user interviews and 23 research rounds.
Proptech had no AI product to look at. We had to invent the interaction language.
BuildingMinds manages real estate portfolios for some of Europe’s largest asset owners. When AI entered the picture, there was no established UX pattern for conversational analytics in enterprise proptech. No competitors to study, no playbook to follow.
I led the design from scratch: research methodology, interaction principles, AI UX patterns, and the design system that made it all consistent and scalable across a team of five. 57 user interviews, 23 research rounds. One guiding principle: make the complexity accessible.
The work wasn’t greenfield. It meant migrating an existing user base from a rigid, tab-heavy legacy interface (built for compliance, not for people) to a responsive, contextual UI language that could carry both structured data and conversational AI without contradiction.
A complete portfolio intelligence layer: ESG reporting, retrofit planning, data management unified under one interface.
The platform spans multiple domains: sustainability compliance, energy modelling, retrofit scenario planning, invoice extraction, asset data collection. Each one dense with data. The design challenge was preventing fragmentation: making the whole feel coherent when the parts were genuinely complex.
Two AI interactions. One had to feel like a conversation.
The other had to show its work.
Conversational AI. The interface had to feel like a knowledgeable colleague, not a system to operate. Prompt examples, a skill selector, and conversational framing were designed to remove the cost of the first message, making an open-ended interface feel safe to start.
AI in forms. When Proxima extracts a value from a document, the pencil icon locates it in the original source: invoice, certificate, scan. You see exactly where the number came from. That’s the control mechanism: not an override, but a verification.
Numbers that came from real measurement, not estimates.
Every metric was tracked across user testing sessions and post-launch analytics. The redesign was not incremental. It was a ground-up rethink of how property intelligence should work. It went on to earn an iF Design Award 2026 in Product Design and a UX Design Awards 2026 nomination.
