Software engineer working on LLM inference and training. M.Sc. Computer Science @ RWTH Aachen.
Active work on git.kokoham.com.
- Multi-Token Prediction (MTP) and speculative decoding for local inference
- KV cache quantization and Metal GPU kernel optimization
- Diffusion-based training for hybrid attention+linear architectures research
- Ternary weight quantization research
ds4-nvfp4-spark — fork of ds4 adapted to run expert-prunned, mixed NVFP4 quantizations of DeepSeek V4 Flash
omlx — personal fork with MTP decoding and Q4 KV cache with Hadamard rotation
sleepy-llm — Zig-native LLM inference engine with hand-tuned Metal kernels
sleepy-agent — fully local Android AI assistant, on-device Gemma 4 inference
- Cross-compiled the core RASR ASR inference engine from x86 to ARM and Android.
- Built a streaming on-device ASR demo on Pixel 6, benchmarking ONNX Runtime and TFLite as independent backends.
- Trained streaming ASR models in PyTorch with causal topologies and hard latency constraints for edge deployment.
Python · Zig · C/C++ · TypeScript · Kotlin