I interviewed with Nvidia for a Backend Compiler Engineer position after attending their campus day, where I spent 4 hours networking and expressing interest in compiler roles to recruiters. The first round was with a senior manager. We started with topological sort coding question naturally evolved into an engaging conversation about its applications in automatic differentiation and dataflow analysis. The interview ended up to be predominantly theoretical without actual coding, but the technical discussion was deep. In the second round, despite some difficulties due to accents, we covered pointer analysis, dependence analysis, and architectural considerations behind compiler optimizations. I received a rejection email very early the next morning,

My 45-minute virtual screen started with a deep dive into the projects on my résumé, especially anything that involved low-level C++ or GPU work (e.g., a CHIP-8 emulator and a CUDA lab). I was asked to explain design choices, debugging stories, and how I measured correctness and performance. The second half was a code-reading exercise: the interviewer pasted a short C++ snippet with several constructor calls (default, value, copy) and a destructor, plus a deliberate memory-management oversight. I had to walk through main, state which constructor or destructor fired at each line, point out the difference between copy construction and assignment, and note the leaked heap object and uninitialised member. No live coding; it was all verbal reasoning. Overall the interview focused on my ability to reason about object lifetime, memory safety, and how my past projects line up with a compiler/GPU rol