Worked on enhancing tensor compatibility in the ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp repositories by enabling support for non-512-aligned tensors over RPC. Focused on backend development using C++ and advanced memory management techniques, the work involved implementing flexible tensor initialization and precise allocation-size calculations. Introduced new RPC commands and refactored error handling to address edge cases, particularly for quantized tensors, which improved reliability and debuggability. These changes broadened model compatibility and reduced allocation errors, allowing larger and more diverse models to run via RPC. The approach demonstrated careful cross-repo collaboration and adherence to established framework conventions.