Vetavize is a benefits app for veterans working through the VA claims process. The machine-learning specialist we expected never came on, so I stepped in and built the AI strategy myself, working alongside a small team designing the app.
I broke the model’s job into facets, everything the AI would need to learn to do, from basic grammar up to judging how persuasive and complete a claim is. Then I sequenced them: what we needed for the demo, what we needed for the first build, and what could come later.
The longer goal is a model, VASPER, that learns from real outcomes. As claims come back scored by the VA, the system learns what actually works with reviewers, not just what reads well, and gets better at it over time.