PROJECT 05
PROJECT 05
Overview
The submission team responsible for FDA report documentation had an established process, but every step of it was manual. For every report, which could contain upwards of 300 individual citations, a reviewer had to locate each source document by hand and verify its accuracy. Corrections were tracked outside of any centralized system, and getting a clear picture of an overall report's status required piecing information together from multiple places. I was brought in as the lead UX designer on a small developer-led team to design a platform from the ground up that could bring that process into one place, using AI in the background to automate the citation lookup and give reviewers a faster, more reliable way to do their work.
IMPACT & METRICS
Citations per report that previously required manual source document lookup one at a time
Estimated manual lookup time saved per report as minutes per citation became seconds
Projected reduction in overall report review timeline per FDA submission
Shorter submission timelines meant the potential to accelerate drug approval and reduce time to market for new molecules
DESIGN PROCESS
This project had the energy of a startup from day one. I joined a small developer-led team as the sole designer, and rather than waiting for a formal research process, I worked alongside the team to go directly to the source. I met with members of the submission team who worked on these reports daily to understand exactly how the manual process worked from start to finish. What I learned shaped everything that came after. Reports could contain upwards of 300 individual citations, each requiring a reviewer to manually locate its source document, verify its accuracy, and track any corrections outside of a centralized system. Understanding that workflow in detail gave me the foundation I needed to start designing a better path forward.
Defining the problem on this project went beyond the user experience. Because we were handling highly restricted data and integrating AI into the backbone of the platform, security considerations had to be defined early alongside the design. I worked closely with the project lead, who was also functioning as a backend developer, to think through data access, user permissions, and how the AI would interact with sensitive source documents. At the same time I was having a different kind of conversation with the team about metrics. Early on the team was focused on building the product, but I pushed for us to define what success looked like in measurable terms before we had anything physical to show stakeholders. I helped them understand that metrics were not just a nice to have. They were the language of the business and the key to securing phase two funding. Defining that early set the project up for a much stronger foundation.
With a clear picture of the problem and the technical constraints in place, I moved into ideation with the full workflow in mind. The core experience needed to feel simple even though what was happening in the background was complex. A reviewer would upload a report, the AI would scan it and identify every citation, and the reviewer would work through them one by one, passing or failing each based on whether the source document and its location were correct. Alongside the review flow I designed a dashboard that gave users visibility into all of their reports at a glance, with status, pass and fail progress, and the ability to pin reports they needed to return to. One of the early design decisions I made was to build the side navigation to collapse, because having both a report and a source document visible on screen at the same time takes up significant real estate. Getting ahead of that problem in ideation rather than trying to solve it later was important. I also started leading conversations with the development team about what future phases could look like, including the ability to assign citation ranges to multiple reviewers on large reports.
The final design brought together every piece of the workflow into a cohesive platform. The citation summary gave reviewers a full picture of where a report stood at any moment, with a table view of every citation, its status, and any comments attached to it, all exportable to PDF for sending corrections back to the writer. File conflict handling was built into the upload flow so that if a revised report came back without a name change the system would catch it and ask the reviewer whether to override or keep both versions. The report a problem feature was designed not just as a feedback tool but as a diagnostic one, asking reviewers targeted questions about what they were doing and where they were in the process so the team could reproduce issues quickly. I also worked with the team to incorporate a confidence score into the citation review UI, giving reviewers visibility into how certain the AI was about each source match, alongside a thumbs up and thumbs down mechanism to feed accuracy data back into the system over time.
Being part of the conversation early is something I actively seek out, and this project is a strong example of what that can look like in practice. As the sole designer on a developer-led team, I was contributing across design, security thinking, product strategy, and metrics advocacy all at once. That kind of broad involvement is not unusual for me, but on this project it carried particular weight because the team was small and the stakes were high. The startup energy pushed everyone to move quickly and think broadly, and having a designer at the table from the beginning meant those conversations had a user and business lens from the start. One of the things I am most proud of from this project is how early we baked measurement into the platform. Too often metrics are an afterthought. On this project I made sure they were part of the foundation, which gave the team something concrete to bring to stakeholders when the ask for continued funding came. Rolling off before the platform reached users was difficult, but leaving behind a clear MVP and a documented vision for future phases meant the work had a direction to grow into.
CONTINUE EXPLORING