It was remote. I applied on LinkedIn and then got an additional form via an email which I also filled. After a few weeks I had an interview scheduled for the following week. The interview was remote, I was told step one is a remote interview about general skills and the second step would be a more technical interview and, likely, on site. The interview began with some easy Python questions. There were some C++ vs Python memory management questions, but not to great detail. Then the focus shifted solely towards Python. They included garbage collection, differences between tuples, lists, and sets and how they are saved in memory. Afterwards, they started asking more about the neural network part. I got questions about Convolutional Neural Network architecture and standard activation functions. Then they started asking about likelihoods, and to be honest, what was annoying was that they weren't asking complicated things, but they clearly had an idea of what I should answer. This is not a format I like, I would like them to give me a question and allow me to explain it, rather than slowly attempt to guide me towards their wanted answer. The second question was something I liked more, about ADAM working principle and advantages over regular gradient descent. The last question was about gradient backpropagation. Basically, I had to explain a few steps of backpropagation and answer a situation where some weights are much larger than others weights.