UnitPulse Portal

An AI-native leasing platform for property owners and their teams. It finds renters, converts them, and measures the funnel. We built it from zero starting February 2026, and the first paying customer signed a month later. It started as separate products built fast. As design lead, I brought them into one platform with one design system.

Leasing still runs on spreadsheets, 7 disconnected tools, and messages nobody answered. The platform is built around 3 people.

The map below shows the 10 main AI agents across the 5 products, with rules-based steps between them and people taking over where it matters. No agent runs the others. They meet at one shared renter record, so a text, a call and a tour all land in the same profile. I designed how people stay in charge: the AI steps back once a person replies, and Copilot waits for a yes before it changes anything.

Marketing was our first product and our first paying customer. It started as a post generator and grew into an AI marketing team inside the Portal.

UP Leasing turns that demand into signed leases. The AI answers every inquiry knowing who’s asking, writes follow-ups worth reading, and books the tour itself on the on-shift agent’s calendar. When the system doesn’t have the answer, the question goes to an agent as a task.

UP Insight closes the loop for owners who aren’t analysts: the full funnel, per-prospect attribution, the biggest leak called out each morning, and a plain-language Data Copilot. Attribution shows which segments and channels actually sign leases, so the team can decide what Marketing writes next.

When an agent is on a tour, the phone still rings. I designed the setup board, call review, inquiries, work orders and the conversation flows. The AI is the overflow, not the front line.

Leasing happens on foot, so we built the agents’ app from scratch in 2 weeks, and I decided what went in first. The feature I pushed hardest for was tour recording: it summarizes the visit and drafts a follow-up the agent sends, so leasing’s highest-signal moment becomes data instead of a fading memory. Around it sit chat and calls, tour scheduling, prospect management, and the Copilot.

Customers came for Marketing, then asked us to run the whole journey. But the 3 products didn’t talk to each other, even though they shared one backend: 3 logins, 3 user-management systems, no shared front door for the AI. So we merged them behind one login and one design system, with the Copilot as the shared command layer, and rebuilt the navigation so each persona sees what’s theirs without wading through everyone else’s. We merged in phases instead of rewriting, and Voice AI later joined the same shell as the fourth product.

I owned the navigation, access states and interaction rules; engineering led the technical integration. As of October 2026, more than 100 properties run on the merged platform.

The 3 products were built fast, with AI writing Tailwind before AI was good at it, so every screen spoke its own dialect. I wrote the language down once in design.md, built it as components, and moved the Portal onto it. It grew from a first version I built in Claude Design.

The same For You page on the first version of the system, and on the new one.

One property shows the whole funnel: a new 93-unit building, 3 leasing agents, 2 months. Every lead came in during that window, and every signed lease came from those leads.

The building was in lease-up, filling all 93 units at once, so there’s no before to compare against, and I don’t claim the platform caused a lift. What it did was run the whole funnel and show where it leaked. Tour recordings kept surfacing the same objections: no parking, and bedrooms facing the stairwell or the elevator.