Agentic AI for reproducible research.
Continue Lecture 8, “Agentic AI for Environmental Data Science.” Practice giving an agent a bounded task, checking its work against the data, and documenting the research decisions that remain yours.
This week
A bounded task
- Specify the inputs, expected output, and stopping point.
- Begin with a small sample and one inspectable artifact.
- Keep credentials and private information out of shared files.
Verification
- Check observation units, keys, joins, missingness, and row counts.
- Preserve raw data and document every transformation.
- Trace a figure or claim to the code and data that produced it.
Research decisions
- Turn research feedback into a concrete next task.
- Explain assumptions, uncertainty, and limits on interpretation.
- Record decisions, actual checks, and unresolved issues in a lab notebook.
Lecture slides
Research follow-up
Individual Research Kick-off Report feedback and feedback slides will be discussed privately. Use that feedback to choose one feasible research task with explicit inputs, an output you can inspect, and checks you can explain.
The public teaching guides below cover research design, data choices, and implementation methods without student assessments or student-created work:
Synthetic-data practice
The starter uses synthetic smoke and weather data. Inspect the supplied files, explain the intentionally failing audit, and plan a checked join without changing the raw data. These invented values support practice with the workflow, not environmental findings.
By the end of Week 11
Be ready to explain a bounded agent task, the checks used to evaluate its result, and the decisions that require your research judgment. Keep a record of what actually ran, what remains uncertain, and the next step.