How I kept learning while teaching

Two NPTEL courses taken while teaching full time, three years before the next one. If you are wondering whether it is too late to start, this is what starting looked like for me: two courses, eighteen months apart, in subjects I was already expected to teach.

The TensorFlow track, taken in order over about fifteen months. Introduction first, then convolutional networks, then text, then sequences — the order matters more here than in any other track, because each course assumes the one before it. The two custom-training courses at the end are where you stop using the framework and start bending it.

The deep learning specialisation, taken alongside the tool track rather than before it. That was deliberate: I find the mathematics lands better once you have already watched something train badly. Structuring Machine Learning Projects is the one I recommend most often and the one students skip most often.

Fairness, bias, interpretability and transparency, taken as my own research widened from privacy alone towards the broader question of what a model is allowed to do. These are short, and they are the ones I would make compulsory if I could.

One course, taken in 2024, when it became clear that every student project was going to involve a language model whether or not the syllabus said so.

Not courses. The Anna University recognition is what lets me take doctoral candidates; the other is a contribution to an edited volume on assessment in higher education.