The first interview was to send me a data set. Given a set of training cases, you need to extract the features and train a model, to predict for the test set. The whole thing went through email, and I think it's close to a joke. To be fair, the problem is interesting to solve. However, I doubt they really value the thinking process of how you build a model. I said that because when I finished the model, which by my best knowledge, performs good enough after careful feature extraction and model selection on K-fold CV for training set, they told me my predictions are way off. And not a little bit of off, way off. This could be seriously surprising since if the training and testing set are sampled the same way, it shouldn't differ that much because K-fold CV would prevent you from overfitting the training set. So, 1) either it's not from the same distribution 2) or they are just being ridiculous And I found they don't treat your work fair enough and value your thinking process. I write detail comments, but I don't think they really read the code at all. Because if they do, they would know I'm doing a good job setting up a model, and doing feature extractions etc carefully. When I asked why I got rejected right away, the data science team declared I didn't do "feature extraction and plot feature selection to see correlation" etc, which is quite absurd given that's all in the comments. And hey, plotting features? No one still doing feature selection by plotting correlations In summary, they don't value your time (don't look at your code at all) and thus don't value your thinking process. From the response they have, I doubt they have a great data science team yet.