Now
Current work
A short record of where my attention is going, at UC San Diego and on the apps I build. Updated when something material changes.
AI as institutional infrastructure
TritonGPT is available across campus and runs by default on university-hosted infrastructure, so routine campus AI traffic stays on UC-controlled hardware. Voice mode and a coding sandbox are live in the assistant, and it now works as an LTI tool in the learning management system, so courses can use it without a separate login. The campus-wide launch for faculty and staff shipped with a full training program behind it. Earlier this month I delivered the campus leadership update on the citizen developer and sovereign AI pillars, and it turned into a much longer conversation than planned. University-wide sessions follow in November and December, with citizen developers co-presenting.
Governed AI enablement
The TritonAI Harness is out of pilot and is a supported build path for development teams, with a broad rollout and training program now live. The usage model is set: on-premises inference is free for administrative use, while cloud usage requires a funding source. Past 100 active users, so this is a product now, not a pilot.
AI agents in the business of the university
Agents are moving out of the assistant and into core workflows. Two are now live: one drafts compensation offers using peer benchmarks, and students and faculty request their own AI platform access through a self-service agent instead of a ticket queue. The near-term roadmap is locked: an institutional-data Q&A tool now live for leadership, automating cash application in the financial system, and an enterprise data agent demo. Next year's budget planning sharpens all of it. With a structural gap between cost growth and revenue growth, every plan is judged on what it does to labor and cost, not personal productivity.
AI workflow pilots and services
The Student Scheduling Assistant is live for fall planning and seeing record use, alongside a voice assistant in the campus mobile app that answers questions using live university data. Other work includes natural-language access to structured data, automated transfer credit evaluation, academic integrity triage, and student success analytics.
Student system stabilization
The student system held through fall enrollment after a difficult summer, and the class-planning tool has doubled in use as it becomes the go-to path for course planning. The work now is data quality cleanup and deciding the longer-term replacement path, which is firming around that same tool.
Data platform modernization
The lakehouse migration off our legacy data warehouse is in evaluation: the RFP closed with 19 bids, and the field is narrowing toward a hybrid architecture, one platform for data engineering and another for integration and governance. Vendor demos are running, with a decision likely December as pricing converges. The target architecture is firming: data lands in a lakehouse on university-owned cloud storage, and the reporting tools people already use stay as consumers. On the observability side, architecture is locked: open-format telemetry feeds a data lake that vendor tools consume from. UC San Diego owns its telemetry data. Endpoint security is moving to a shared systemwide product next year, with campus migration starting in the new year.
On-prem AI infrastructure and expansion
The shared AI platform now runs assistants for several other campuses, and the newest partner went live in September. Two newer moves: a peer campus will start paying for gateway usage, a first step toward offering the gateway as a service, and a shared GPU cluster for data science and machine learning has received funding and is moving forward, with a smaller shared cluster for teaching and learning alongside it. The data science platform relaunches this month under a new name with refreshed documentation. A pilot systemwide inference service is taking shape: our campus provides the open-weight models, and three partner campuses consume them through our gateway. A small central team to coordinate access to AI platform capabilities is proposed.
Core infrastructure consolidation
We finished moving about 1,500 virtual machines to an alternative hypervisor in roughly seven months, and UC San Diego is now a reference campus for peers working through the same licensing shift. Now the question is the main on-campus data center: the colocation model is set, the first partner racks are arriving this month, but ownership and funding are still to land.
Identity management
Kicked off a project to rebuild the identity registry as the foundation for access management. Starting with the registry itself before tackling downstream automation. Alignment with the health sciences side of campus is a key risk we're managing.
Systemwide and national AI strategy
A systemwide AI council is now negotiating systemwide licensing directly with the major model providers, and I'm pushing API-first access over single-vendor lock-in. Contracts are expected soon, with final decisions landing with the Regents by December. Our campus AI strategy goes to the Regents this month. Separately, our campus is the on-premises and open-source tier of that strategy. I also joined the NASH AI Advisory Board, the first national AI effort built exclusively for public higher education systems, on its Data Infrastructure & Privacy working group. I'm co-writing an article for a UC technology publication on campuses sharing one open-weight platform, and a recorded podcast conversation on the same theme is coming.
The apps I build
Outside the day job I build iOS apps. Resolution Companion 1.6 is live on the App Store, adding per-habit reminders. Horse Racing Companion is live, with the fall meet opening in early November. Steel City Gameday is mid-season with its public Crew feature live, and a fix for cross-game replies shipped this week.