uptime

Sole IC across the entire data + AI platform. Scope that's typically distributed across 4-5 engineers (data engineering, ML engineering, analytics, platform ops, governance) owned end-to-end as one role.

The work breaks down:

AI capability: Six AI systems across the platform. Leadership-facing NL analytics on Azure OpenAI GPT-4o, indexed over multi-million-row registration data, with 40+ FastAPI endpoints. Conversational analytics agent with function calling and audit logging. Edge AI occupancy monitoring (Genetec KiwiVision, >99% accuracy, live since January 2026). Vector-search event chatbot in production. Automated content-review pipeline at recurring cadence. Local-LLM knowledge-base RAG agent.

Data fabric: Production lakehouse on Databricks. 30+ daily pipelines integrating every operational source system across the org. Self-service analytics across six departments on a consolidated query layer.

Governance + security: Authored the AI policy the org operates under. Custodian of the $80K annual POC budget. Renews each year, every system on it still in production. Sole admin of the GitHub org. Security audits caught critical vulnerabilities in vendor-delivered code.

Current commitment: A Customer 360 initiative. 13-capability platform deliverable to the board for November 2026. Phase 2 and Phase 3 mapped.

Program Intelligence runs Azure OpenAI GPT-4o over the org's full registration history. 2.6 million records indexed, 40+ FastAPI endpoints serving structured queries, function-calling for parameterized drill-downs against the warehouse. Leadership asks natural-language questions (revenue by program, attendance trends, year-over-year shifts in any cut of the data) and the system answers in plain English with citations to the underlying rows. Every conversation is logged with the prompt, the function plan, the SQL the model actually executed against the warehouse. Reviewers can replay any session end-to-end. Adoption: 15+ users on the executive leadership team querying directly. No analyst middle layer between question and answer. Built the data model, the API layer, the agent, and the auth, deployed it under the AI POC budget, kept it in production.

Live occupancy monitoring at the edge. The org needed accurate, real-time headcount across a multi-zone facility, surfaced on operations dashboards in real time. Off-the-shelf cloud inference was a non-starter on bandwidth and latency at the deployment sites, plus accuracy under variable lighting and crowd density was a real problem. Wired in Genetec KiwiVision (video-analytics platform with on-device inference) running models that landed >99% accuracy on the target classes. The integration produces alerts wired into the existing operations dashboards. Health checks on the cameras themselves: a dropped feed pages before anyone notices visually. Live since January 2026, unattended through hardware refreshes and zone layout changes. Operations staff get the live signal they were doing manually. The deployment was budget-line under the annual AI POC budget.