VA AI Copilot

The hand I was dealt

The VA wanted an LLM, similar to ChatGPT but built for VA employees, to cut time spent on administrative work like drafting text, writing emails, and analyzing documents.

A few weeks in, the agency went through a major reorganization and workforce reductions. New leadership in the Office of the Chief AI Officer pivoted the tool toward clinicians: primary care physicians, nurses, and their teams.

They spend a lot of time sorting through volumes of patient information that's often cluttered, poorly labeled, and takes multiple clicks to reach and search for. If the clinical rollout succeeded, then the tool would be expanded to other teams.

That left us with two problems:

• First, trust: after the layoffs, staff saw AI as a possible threat to their jobs, so how do we make it a tool they'd trust and want to use?
• Second, the actual use case: no one had defined how physicians would actually use it day to day.

To maintain the confidentiality of the AI Assistant’s product name and not violate NDAs, the term “VA GPT” will be used.

My contributions

• Led initial design direction and enhanced visual styling for improved accessibility
• Collaborated within a cross-functional team to create a usable LLM interface
• Conducted user research and testing for the LLM MVP release
• Owned creating a collaborative baseline design system for the LLM
• Improved user flows and UI design to increase document-upload efficiency, citation accuracy, and PII security