Generative AI in A+ lets brands create product copy and lifestyle imagery in a fraction of the time it used to take them. The beta made the demand obvious: 276 brands signed up, and 40% of them were using it within the first seven weeks. My job was to take that promising prototype, design the production experience, and lead the team that shipped it.
I owned the front-end design and led a team of 4 engineers against a fixed launch window. I wrote the design doc with near-complete pseudocode for each task so the team could execute in parallel without re-deciding the approach as they went. This cut our implementation time by an estimated 20%.
Key Contributions
Product & UX
- Designed the prompt experience that gives brands real control over the output: they can feature a product or not, select or upload a custom reference image, choose a realistic or abstract background, and describe the setting in their own words.
- Ramped the feature out by allowlisting module types as generation quality cleared the bar, reaching 70% coverage of active modules before fully opening it up.
Architecture
- Since image and text generation are long-running requests, I introduced an async polling pattern the team hadn't used before, then documented it with a sequence diagram and pseudocode. That doc is now the team's reference for async API integration.
- Built the front-end against an API contract that was still evolving, and kept the client aligned with the back-end as the generation service matured.
Measurement
- Built the metrics dashboards that track text adoption, optional prompt-field usage, and total generation volume, then a second set for the international expansion.
Result
We shipped for the Amazon Accelerate 2024 announcement to all US users, then expanded to 18 marketplaces around the world. The feature now runs at about 130,000 generations per month with a 91% adoption rate on generated text, and it was recognized on stage at Amazon Accelerate 2025.