The operating environment
I owned the production path behind CS 220's AI tutor, autograder, analytics, and assignment releases.
PRODUCTION AI INFRASTRUCTURE · SECURITY
CS 220 AI Learning Infrastructure
CS 220 serves more than 900 students, and deadline days turn small inefficiencies into operational problems. I led a 12-person team and rebuilt the tutor, grader, and observability around the moments the course was under the most pressure.
900+ students on autoscaling Azure infrastructure
80% lower LLM inference cost
12× faster grading · 12K+ submissions
50% faster incident response
Rebuilding the AI tutor for course scale
Rebuilt a single-threaded Flask service as async FastAPI on Azure Container Apps, so replicas scale with demand.
Reducing LLM cost without reducing access
Gateway controls kept student access fast while repeat and unauthorized traffic stopped reaching the paid provider.
Closing the autograder trust-boundary failure
Student notebooks ran as root beside hidden tests and Canvas credentials, so I rebuilt the grader around an OS privilege boundary.
Making failures visible
Instrumented the services so staff could go from a failure report to the responsible request.
Releasing without disrupting the course
Moved releases into a GitLab CI/CD pipeline with a blue-green cutover, so deadlines never hit a maintenance window.