Suh worked on the NVIDIA/NeMo repository, focusing on backend stability and correctness in audio processing and speech-to-text services. Over two months, Suh addressed critical bugs in the AudioToMelSpectrogramPreprocessor, ensuring tensors were placed on the correct device and shaped as expected, which improved GPU pipeline reliability. In addition, Suh enhanced the NemoSTTService by correcting model type checks, fixing missing argument calls, and updating documentation to reduce runtime errors and improve maintainability. Using Python, PyTorch, and testing frameworks, Suh’s contributions deepened test coverage and strengthened production readiness, demonstrating a methodical approach to backend development and machine learning workflows.