AI in Research and Teaching

Oct 9, 2026 3 min read

This is a living page, collected as we go, for anyone who uses AI in research. It will grow, and some items will be reviewed or removed over time.

Listing a resource here does not mean that we endorse it or its authors. These are materials we found interesting, shared for information; please use your own judgement and check any tool, policy or claim before relying on it.

✅ Good use

What works well in research, what to avoid, and how to stay transparent about it.

To be added.

🎓 Teaching

Ideas for teaching students to use AI responsibly: verifying outputs, reproducibility, authorship and disclosure.

  • Eight Rules for Teaching in AI World (Kevin A. Bryan, University of Toronto, August 2026): an essay arguing that AI breaks the link between assignment performance and learning, with eight rules for redesigning courses: decide which skills must be shown without AI, avoid take-home work that AI can complete, give students incentives to study, and state a clear AI-use policy (an example is included). The author founded All Day TA, an AI teaching company that the page uses as its main example.

🧰 Tools and skills

Useful tools, and reusable instructions (“skills”) for Claude and other assistants that save time and improve quality.

Practical AI for Academics (Claes Bäckman, Aarhus, 27-28 August 2026): a two-day workshop with open materials, including slides, setup guides, written guides on agentic AI for empirical research and on verifying LLM output, and a CLAUDE.md starter template. The Claude Code skills (MIT licence) are on GitHub, among them a referee-style review of your own draft and a skill that builds a voice.md style file from your own writing. Written for economists, but the workflow carries over to epidemiology.

Paul Goldsmith-Pinkham, series on AI coding tools for empirical research (Claude Code). Written for economists, but the ideas carry over to epidemiology:

Repositories of skills and templates for Claude Code:

  • Claude Code Academic Workflow (Pedro H. C. Sant’Anna, Emory University; GitHub template, MIT licence): a full setup for academic research, mainly economics, with skills (papers, data analysis, replication packages, slides, teaching), agents (referees, proofreaders, claim verifier), rules, hooks and a command that runs a set of checks. It is a large setup, so adopt only the parts you need.

  • MixtapeTools (Scott Cunningham, Baylor University): tools for coding, teaching and presentations with AI assistance, framing Claude as a thinking partner rather than an autonomous agent. It includes audit workflows (a replication-style audit run in a fresh session, a review for overlooked problems and a bibliography checker), slide-making workflows and a CLAUDE.md template. It states “use freely”, with no formal licence.

  • ChernyCode (meleantonio, MIT licence): a general software-development template that turns the Claude Code tips of Boris Cherny, its creator, into ready-to-copy configurations: project memory files, skills (such as commit, testing and code simplification) and subagents (such as code reviewer and verifier). It is not aimed at research.

🔭 Potential interest (not yet curated)

Material collected but not yet reviewed. Treat these items as leads, not recommendations.

  • To be added.
Otavio Ranzani
Otavio Ranzani
Group Leader

I am a scientist, born in 🇧🇷.