Jovian is a small company with two genuinely good courses and a business model that has never quite resolved. I want to separate those two things, because the courses deserve a recommendation and the company deserves caution, and conflating them would do a reader a disservice. Start with the teaching, which is the reason to be here. Deep Learning with PyTorch: Zero to GANs is, in my view, one of the better free introductions to PyTorch in existence, and I do not say that lightly given the competition includes PyTorch's own tutorials and Daniel Bourke's course.
What it does well is refuse to let you be a spectator. The structure is lesson, then assignment, then the next lesson, and the assignments are graded. You cannot drift through it with a video playing in another tab, because within the first week you are being asked to implement linear regression with tensors and getting told when you are wrong. The progression is sensible too.
Tensors and gradients, then linear regression, then logistic regression on MNIST, then feedforward networks, then convolutional networks on CIFAR, then generative adversarial networks at the end as the payoff. By the time you reach GANs you have built up every component honestly, which is exactly how this material should be sequenced and frequently is not. Data Analysis with Python: Zero to Pandas is the other pillar and has drawn something in the region of eighty five thousand learners. It is more conventional in scope, covering Python basics, NumPy, pandas and visualisation, but it carries the same insistence on doing rather than watching.
The course project requirement is the part I would single out. To finish you have to find your own dataset, analyse it and publish the notebook. That is a small thing that produces a disproportionate result, because it forces the transition from following instructions to making decisions, which is the transition that most beginners never make and the one that separates people who finished a course from people who can do the work. The whole thing runs in hosted notebooks, so there is no environment setup, which matters more than it sounds.
Now the company. Jovian started as a notebook sharing and versioning tool for data scientists, something like a specialised alternative to sharing files around, and the courses were partly a way to build an audience for that product. Over the years the emphasis has shifted repeatedly. There have been paid bootcamp programmes, a Zero to Data Analyst bootcamp, a Zero to Data Science bootcamp, various career oriented offerings at real prices, and more recently the platform has visibly been reworked again.
Some parts of the site feel actively developed and some feel abandoned in place. The live cohort runs, which were the thing that made the courses feel like a class rather than a playlist, have not run consistently. That inconsistency is the honest reservation. If you sign up expecting a supported cohort with instructors answering questions in a forum, you may find yourself working through recorded material alone.
That is still worthwhile, because the material stands up on its own, but it is not what the framing implies and you should go in knowing it. The certificate is worth nothing in the market. I say that without malice, because it is true of nearly every completion certificate and Jovian is not unusual. Nobody in hiring knows what Jovian is.
What you get out of it that is worth something is the project notebook you published and your ability to talk through the decisions in it. The course knows this, which is why it makes you build one. There is also a scope limit worth naming. These courses take you to competence in the fundamentals and then stop.
There is nothing meaningful about deploying a model, about evaluation beyond accuracy, about data pipelines, monitoring or the parts of the work that occupy most of a practising engineer's time. Zero to GANs will teach you to train a convolutional network and will not teach you what to do with one afterwards. That is a legitimate design choice for an introductory course, but it means Jovian is a first step and cannot be a whole path. My 3.5 balances a genuinely strong pedagogy against a shaky institutional foundation and some ageing edges.
The deep learning course in particular punches far above the visibility of the company that made it, and if PyTorch has not clicked for you yet I would try this before almost anything else free. But I would take it for the understanding and the project, not because you expect Jovian to be a place you keep coming back to. Treat it as two very good courses that happen to share a website. On that basis it is easy to recommend.