Designing trust into an AI-driven recommendation. The central risk in an AI-powered home matching product is that users don't trust the output — especially when the stakes are as high as a home purchase. An unexplained recommendation, no matter how accurate, creates anxiety rather than confidence. The critical design decision was to make the reasoning behind every recommendation visible: surfacing which lifestyle signals drove a property's match percentage, and which factors were creating misalignment. This transparency wasn't just a UX feature — it was the mechanism that made the AI feel like a decision-support tool rather than a black box.

Converting qualitative lifestyle preferences into structured data without losing nuance. The core technical challenge was that the signals that actually drive good home decisions — "I want to live somewhere walkable but quiet," "we have two dogs and a toddler" — are inherently qualitative and conversational. The design decision was to build a conversational onboarding model that gathered these signals in natural language and translated them into a structured profile behind the scenes, rather than forcing users into static filter inputs that would strip the nuance out. The constraint this solved was that traditional search UI (beds/baths/price) systematically excludes the signals that most predict long-term satisfaction.

Designing for real estate agents as a parallel user, not an afterthought. Early in the project it became clear that buyers rarely navigate high-stakes home decisions alone — agents are almost always in the loop. Rather than designing purely for the end buyer and hoping agents would adapt, I architected the information model to be legible and useful to agents reviewing a client's HomeMatch profile. This meant the system needed to communicate not just recommendations, but the reasoning behind them in language an agent could use in a conversation with their client.