Field notes from an active researcher integrating generative AI into teaching, proposals, code, and mentoring — documenting what works, what fails, and what I haven't resolved.
Posting format: each post is a single markdown page, and can include a link to the public GitHub repository containing skills/instructions files.
Community note: to comment, open a post and click Discuss this post →. Discussions happen on GitHub to keep the conversation open, searchable, and collegial.
I rebuilt my machine learning for geoscience course in short morning: agents, evaluations, twelve adversarial reviewer personas, and translations into French and Spanish. The scarce input was deciding who the course is for.
Retrieval tooling has made agents genuinely good at research. The price of admission is full transparency — not just of data and code, but of reasoning.
A postdoc’s workflow with Obsidian, Claude, and VaultNotes suggests a new loop from speech to planning to scientific execution.
A Warning Back in 2025
Sixty-Four PDFs and No TA
What I Should Have Written First
One Prompt, Two Deliverables, and a Sign Error
The Slide That Started This