hi, i'm emre — ml systems & performance engineer.
i build and optimize inference systems for production: low-latency audio models, custom CUDA/CuTe-DSL kernels, and the surrounding infrastructure that makes them ship.
recent work:
- shipped 430ms → 50ms STT latency in production (8.6x) via FP8 quantization, fused kernels, and CUTLASS EVT epilogues
- contributed Consumer Blackwell (SM120/SM121) support to Dao-AILab/quack — RMSNorm, softmax, cross-entropy kernels in CuTe-DSL (merged)
- merged PR on NVIDIA/CUTLASS: Snake activation as an EVT op (2.1x median, 8.5x peak on Freya-TTS)
focus: GPU kernel engineering, inference latency, and production deep learning systems.