MLOps mentor

Goal

Four hundred and fifty students doing course projects deserve feedback while they work, not after grades land. No teaching team can read that many repositories by hand, so the reading itself had to be automated.

Report

Built rasgaard/mlops-mentor: repo mining for concrete signals (CI setup, dependency files, code volume) plus a language-model pass for qualitative feedback. Results flow into a public dataset and a Gradio app anyone can browse.

Two honest footnotes. It never quite became an official course tool; the hope was to shape it together with the students instead. And TA-ing made something obvious: language models have absorbed the TA’s old role as the patient, always-on helper. What a course like this really teaches now is stamina, the art of wrestling dependency hell until things just work.