Developed initial Vulkan-based IQ2 and IQ3 quantization support for both Mintplex-Labs/whisper.cpp and ggml-org/llama.cpp, focusing on enabling faster and more memory-efficient inference on Vulkan-enabled hardware. The work involved creating new shader pipelines and dequantization routines in C++ and GLSL, supporting multiple quantization variants such as IQ2_XXS, IQ2_XS, IQ2_S, IQ3_XXS, and IQ3_S. By aligning code and optimizing for Q3_K quantization, the developer established consistent APIs and integration points across both repositories. These enhancements expanded hardware compatibility and laid the groundwork for improved performance in quantized model inference using GPU computing techniques.