I applied online. The process took one week.

There was just an OA. The whole thing was 70 minutes long. It started with seven multiple-choice questions (not too bad), one LeetCode medium, and two ML from-scratch coding problems. But that coding part took me a solid 50 minutes just struggling with the tool constraints.

The multiple-choice questions covered precision and recall calculations, the differences between bagging and boosting, overfitting-related concepts, and manual forward/backward propagation calculations for neural networks. However, the programming section was highly restrictive. We were limited to using scipy.cdist for KNN and sklearn.tree.DecisionTreeClassifier for decision trees. As a result, I spent the majority of the time checking the documentation. I couldn't understand the restriction on NumPy, as many of my standard NumPy operations were unavailable, and I couldn't come up with alternatives under the time pressure. Consequently, I left the final three problems incomplete.