LinkedIn Senior Machine Learning Engineer Interview Experience — Debugging Logistic Regression, Then Job Click Prediction Design

The first part was 30 minutes of debugging a logistic regression implementation. I spent too much time on the sigmoid derivative at the start, so I didn't finish the last step, computing the gradients.

Then came 20 minutes of ML design: Predict whether a member will click on a displayed job using member features, job content, and interaction history.

I had to define everything myself: what data is needed, historical interaction data, system components, how to train, the loss function, how to prevent data leakage, what the supervision signal is, and how to get user and item embeddings.