Scale AI New Grad ML Research Engineer Onsite

Four interviews happened on the same day: ML fundamentals, an LLM practical, behavioral questions, and coding.

The ML round covered NaN loss, CNNs, differences between computer vision tasks, and a paper about masked attention.

The practical was Training Data And Model Error Debugging. It started with reading JSON files from a directory and writing the data to CSV. After that, they had to fix three bugs. The browser editor slowed down during the task.

The coding round started with Shortest Distance Between Two Nodes, then moved to Minimum Distance Matching in a Forest. The input was a nested dictionary with values and child lists, wrapped in a top-level list without one root. They discussed lowest common ancestor and another offline approach.