NPTEL is one of the largest and least discussed online education efforts anywhere. It is funded by the Indian government, delivered by faculty at the IITs and IISc, and it has run for well over a decade producing thousands of courses across engineering and science. Outside India almost nobody has heard of it, which is a shame, because a lot of what is on there is better than what people pay significant money for elsewhere. What you get is a real university course.
Recorded lectures from an actual IIT professor, assignments, and a proctored exam at a physical centre if you want the certificate. The mathematics is not softened. Where a typical online machine learning course will gesture at the derivation and move to code, an NPTEL course will work through it on a board. If you want to genuinely understand why an algorithm works rather than how to call it, this depth is hard to find at this price.
The catalogue is remarkable. Beyond the standard machine learning and deep learning offerings there are courses on reinforcement learning, computer vision, natural language processing, optimisation, pattern recognition, probabilistic graphical models and a long tail of specialised topics that simply do not exist elsewhere as free structured courses. Balaraman Ravindran's reinforcement learning material and the various IIT Madras and IIT Kharagpur machine learning courses are the ones I have found most worthwhile. The certificate matters in context.
For Indian students and job seekers, an NPTEL certificate from an IIT course is recognised, appears on CVs, and in some cases carries academic credit transfer. The exam is proctored at a physical centre, which means it is not trivially gamed, and the cost is nominal. For a fraction of what a Coursera specialisation costs you get something with more institutional weight in that market. Now the things that will frustrate you.
The production is basic in a way that takes adjustment. Fixed camera, a lecturer at a board or reading slides, poor audio in places, no editing. There is no instructional design here. Someone accustomed to a well produced modern online course will find the first few lectures hard going.
Instructor quality varies more than on any other platform I know. Some IIT professors are outstanding teachers who explain difficult material with real clarity. Others read their slides in a monotone for fifty minutes. The platform gives you almost no way to tell which you are getting before you commit, and course ratings are not a reliable guide.
My advice is to watch two lectures before deciding, and to search for the instructor's name and reputation first. The practical gap is real. Most courses are theory and mathematics with assignments that are largely problem sets. There is little coding, few projects, and almost nothing that produces a portfolio artefact.
You can finish a strong NPTEL machine learning course with a solid grasp of the mathematics and no ability to build anything. That needs pairing with a hands on course or your own project work. The schedule is fixed. Courses run on semester dates with assignment deadlines and a single exam window.
Archived material is available to watch any time, and the certification path requires enrolling in a live run. That rigidity is unusual now and it does enforce pace, which some people need. The platform itself is clunky. Search is poor, navigation is dated, and finding what you want across thousands of courses is harder than it should be.
My three point eight is for genuinely rigorous university teaching, free, with a cheap and credible certificate and mathematical depth that is very hard to find at any price, marked down for inconsistent instruction, minimal practical work, dated production and a platform that makes finding the good courses unnecessarily difficult. Pick the instructor carefully and it is one of the best value propositions in online education.