About Core Velocity Lab

Ervin Tasnadi

Independent AI Systems Engineer

Hi, I’m Ervin Tasnadi, an independent AI systems engineer and developer with 10 years of experience in GPU computing.

Core Velocity Lab was founded in 2026 as the formal corporate entity to provide high-end engineering consulting and R&D services for enterprises and technology companies looking to maximize their AI compute efficiency.

Core technologies

• CUDA, CUTLASS, CuTE
• Triton, MLIR infrastructure
• Vulkan, GLSL

Track record

• Efficient implementation of CONV2D in llama.cpp.
• Efficient implementation of CONV_TRANSPOSE_1D in llama.cpp.
• Vulkan implementation of FlashAttention-2 (forward pass): supporting at-the-edge inference on many devices.