Different problems,
different shapes.

In late February I moved to Amazon's autonomous delivery robot team as the first and only UX designer across the entire ADR ecosystem. My job is to shape a cohesive experience that spans multiple systems and teams — simulation, teleoperation, fleet management, last-mile integration, and next-generation human-robot interaction.

There was no established UX practice and no dedicated PM. I work directly with engineering — iterating, discussing, exploring, and evolving together at the pace the problem demands.

Across teams, I coordinate stakeholders and UX designers from different parts of the last-mile system to make sure robotic integration doesn't happen in isolation. The robot touches delivery workflows, driver tools, and customer-facing status — those seams need someone watching them.

On the research side, I'm exploring human-robot interaction for the next generation of delivery robots. How should a person on a sidewalk understand what the robot intends to do? How do remote operators build calibrated trust — enough to let the system work, not so much that they miss failures?

These questions sit at the intersection I keep returning to: physical operations, AI, and trust. Interfaces where decisions have consequences beyond the screen.

Gemini Hackathon project built with a fellow UX designer. We wanted to explore what AI conversations look like when they stop being linear chats and start behaving like spatial thinking — the way ideas actually form when you map them out.

The system analyzes chat history and organizes it into "Objects" — anchor topics of your thinking. Each Object contains "Nodes" — milestones, pivot points, facts, or concepts. As conversations continue, new connections emerge — the way our minds actually work.

As two UX designers without coding backgrounds, we vibe-coded the entire thing with Gemini as our development partner. Followed our UX process — user flows, screens, iteration — but with AI building alongside us. The biggest learning: prompt engineering itself can be AI-assisted. When stuck, we asked AI how to better talk to AI.

The team was building a Business Health Dashboard for Delivery Service Partners (DSPs) — the independent businesses that operate Amazon's last-mile delivery fleet. The assumption was that DSP owners needed a better data dashboard to track business performance.

DSP owners didn't want another dashboard full of metrics. They wanted intelligent guidance — help understanding what the numbers meant for their business and what actions to take. The gap wasn't missing data. It was missing interpretation.

I produced a gap analysis comparing the PRD assumptions against actual user needs and distributed it to stakeholders. Rather than quietly adjusting my designs to compensate, I pushed the conversation upstream — making the case that we were solving the wrong problem and needed to reframe before investing in execution.

The research shifted the product conversation from "what metrics should we show" to "how do we help DSP owners make better business decisions."

Real-time container movement dashboard for Amazon sort center managers. The data was comprehensive but illegible. Ruthless information hierarchy — stripped to what drives the next decision.

A 0-to-1 financial management tool for Amazon construction projects. Worked with another UX designer to establish the information architecture, key modules, end-to-end user flows, and product direction from scratch.

Translated complex financial workflows into clear, navigable structures that construction project managers could use without specialized training.

Sole designer. 70% of smart purifier users never connected to the app. Reframed what the app was for — not more control, but contextual intelligence the physical device alone couldn't provide.

Shipped full IoT experience: setup, controls, settings, automation. Changed the team's mental model of "smart" from adding control to reducing the need for it.

View product →