Pruduct Designer

Machine Learning

I spent some time on our internal Visual Design team, where I crafted a fully customizable system of sticker-like icons that highlighted various regions of a car. The purpose was to give designers a flexible system to show where blemishes were located on the car, but the result was applicable across many use-cases.

Car Part Sticker System

My goal was to imagine the future of EVs at CarMax. I was tasked with aligning with set company goals, reviewing available research, discovering problems and opportunities, and depicting this future to inspire leadership and create alignment.

* I partnered with designers across the organization to create a large, detailed, blueprint of the end-to-end customer journey, as well as identifying pain points, both for the customer, and for the business, at every turn.

* I used generative AI to rapidly create consistently-styled sketches to help tell the story of a present and future customer experiences.

* I created many design concepts that addressed problems and opportunities identified in this exercise, to help that future feel more real.

The Future of EVs

A one hour class to teach designers how to use LLMs with activities reminiscent of design school. The goal is to build exposure and familiarity among designers through a crash course of different AI tools and capabilities.

Prompt School

Problem: There are many types of damages a used car may have, yet we can only feasibly capture, process, and present to customers a small number of them.\
Research Question: What types and locations mattered most to customers?\
Methods:

* Use card-sorting to first understand customers' terminology and taxonomy of car damages.

* A survey asking customers to identify the severity and importance of various blemishes.

Vehicle Blemish Discovery

Code-named Tapas, a component that allowed users to explore CarMax inventory while also getting deep into car details. This was an experimental approach to engaging users coming from search ads with our inventory.

Heads-up Recommendations

I am currently experimenting with designed and testing a multi-agent LLM prompt architecture to bring text generation to our chatbot in a safe, efficient, customer-friendly, and on-brand way.

Prompt Architecture for Chatbots

During my first week on a new product team, I created a vision board for future of CarMax's chatbot experience, Skye. My goal was to imagine the best chat experience in the world. My task was to visually showcase as many unique and innovative ideas as I could come up with in one week. The images showcase what I came up with.

Concept Wall - Future of Chat

My goal was to test the immediate impact of brand photography across the CarMax site. I was tasked with testing different pre-vetted photos on various pages to find which had greatest emotional impact (sometimes referred to as a "visceral" experience).

My approach was to conduct 5-second tests with candidates recruited on UserTesting.com. Each tester gave feedback one version of a photo from each page, and gave feedback on each.

I used an LLM to aid in the synthesis of my research by generating keywords that summarize the feelings and thoughts expressed by all users for a given photo, as well as a by scoring the emotional valence of each. This was used as a loose proxy for experience alongside my own manual synthesis (affinity clustering, summarizing, etc).

Photos Emotional Impact Test

Led the successful redesign of CarMax's web chat and support bot.

Chatbot Redesign

I redesigned the landing page that people would see when they click on one of CarMax's car ads.

Jason Marsh

Work Experience

Certifications

Contact