Applications, AI agents, evaluation research, and the engineering decisions that connect them to real workflows.
AI feedback inside Canvas
I build parts of HelpMe that connect AI feedback to real course workflows. My work includes Canvas LTI deep linking, structured feedback across services, and configurable course agents for LANTERN.
Read the storyInstructor Choose a question
Student Receive feedback
Testing tutoring integrity under pressure
I helped build AiTutor and created a harness to test when tutoring models give away answers under repeated student pressure. That work led to two first-author computing-education papers.
Read the story
From onboarding spreadsheet to SaaS configuration
At DxFactor, I built an agent that turned onboarding spreadsheets into dependent API operations, with retrieved documentation, persistent state, validation, and retries.
Read the storyUnderstand the onboarding data.
Retrieve the API docs for this task.
Use authenticated tools and saved state.
Retry, resume, or report the failure.
Architecture and technical leadership
I was a secondary technical lead on EduAI. I handled architectural decisions, reviewed pull requests, ran daily stand-ups, and held review meetings with professors and other stakeholders.
Read the storyA settings route checked a person's platform role, but not whether they taught the course they were changing.
Check access to the target course before reading or changing its settings.
Some tools for my own development workflow, and things I built for people close to me.
Streams large macOS logs while keeping memory use bounded.
Runs isolated Codex desktop profiles with a local dashboard for account and app-instance management.
Lets a coding agent pause for a person's answer and continue with that decision in context.
Finds a previously visited page using local browser history and on-device search.
A website and accessible interactive activities built for a family mental-health practice.
