Worked on enhancing the stability and reliability of GPU-accelerated data handling in the FAIR-Chem/fairchem repository, focusing on the CUDA data path for atomic structure conversion. Addressed a critical bug by ensuring that AtomicData is moved from GPU to CPU before converting to ASE Atoms, which resolved dtype and device errors in Python-based workflows. Introduced a regression test to validate the corrected data movement and prevent future issues, improving test coverage and error traceability. Leveraged skills in data conversion, GPU computing, and testing to reduce runtime errors and support more robust production-like workflows involving CUDA and scientific data processing.