09/02/2026 updated
100 % available
AI/ML Engineer: LLM inference optimization on NVIDIA GPUs (vLLM, CUDA, PyTorch)
Suwon, South Korea
Only remote
LLMComputer VisionBenchmarkingNvidia CUDALinuxPython (Programming Language)Machine LearningPyTorchDeep LearningFastAPIDocker
Conatus AI is a one-person engineering practice focused on inference performance on NVIDIA hardware: serving stacks (vLLM, SGLang, llama.cpp), quantization (FP8, NVFP4, GGUF), KV cache and memory sizing, and throughput and latency benchmarking on Blackwell (RTX PRO 6000, DGX Spark) and Hopper. Typical engagements: capacity and feasibility checks before hardware purchase, measured benchmark reports with full configuration tables, deployment fixes when a model does not fit or a server stalls, and PyTorch or CUDA level optimization of custom kernels and data pipelines. Background: about 12 years of programming and 8 years in machine learning, ex Samsung Research AI service lab, current research work in efficient inference. Public evidence: technical write-ups at conatus.jahn.ai/ai-engineering and issue-level contributions in vLLM and llama.cpp.
Working mode: written briefs only (email or platform chat), fixed scope agreed in writing before start, deliverables as documents, scripts, or pull requests with reproducible measurements.
Experience
Professional experience10+ years
Freelancer since2026
Languages
EnglishFluentKoreanNative speaker