Work
Uber
Summer 2026- Support for customers who pay to never think about a rideI came into Uber Elite wanting to automate support. Interviews with the executive assistants who book most Elite rides changed my mind, and the future-state model became concierge support built on reaching out before something goes wrong.Summer 2026
- Merging two support models before an acquisition closedUber and Blacklane each wrote a plan for post-close support that defended how they already worked. I got both teams to edit a shared comparison instead of arguing over documents, which narrowed the fight to two decisions.Summer 2026
- When the support tags were wrongWhile mapping where support was failing premium riders, I noticed the ticket tags didn't match what riders were writing. A clustering pass and a hand-coded sample showed four contact drivers were systematically mislabeled.Summer 2026
Kellogg
2025 to presentEY
2021 to 2025- Turning an AI wishlist into funding decisionsAn automotive client's AI program had more ideas than budget and no way to compare them. I rebuilt intake around the three inputs its SVPs actually decided on, and the first review cycle reached decisions in about two weeks instead of five.2024 to 2025
- Getting an AI pilot past a privacy review boardA client's AI Review Board blocked an HR timesheet agent on privacy grounds. I found the one risk the board actually cared about and wrote a narrow fix for it, and the board approved the pilot.2024 to 2025
- Finding where AI actually helped a deal teamA private equity deal team wanted to use AI but didn't know where it would help. I spent two weeks working their hours, found the bottleneck in the time data, and built the roadmap around it. Review time dropped 17%.2024 to 2025
- Missing a release, then fixing how the team plannedI led delivery on a global drug forecasting platform and let a new vendor team skip estimation to keep the relationship easy. We missed a release. Here's what I changed, and how the next major release launched in 70+ markets on time.2023 to 2024
- The $30M that looked like fraudA health system that also runs an insurance plan was paying out far more than its clinical volume could explain. Leadership assumed fraud. The SQL logs showed a date-format mismatch that made the system pay for the same procedure twice.2021 to 2023
- Predicting who drops out of behavioral health treatmentA state health department was spending heavily on beds and still losing patients from treatment. Clinical data couldn't predict who would drop out, so I added census data, built a risk model in Python, and paired it with recommended next steps.2021 to 2023