Huiseom Kim

Huiseom Kim (김희섬) — ML Engineer (6 YOE) I am a Machine Learning Engineer with 6 years of experience covering the full AI lifecycle—from data collection and preprocessing to model training, optimization, and production deployment. I have hands-on experience building real-world ML systems, including a reinforcement learning–based dispatch engine for a last-mile delivery platform. I designed and implemented a simulation environment to train RL agents, enabling data-driven decision making in production. My core expertise lies in NLP and search/recommendation systems. I have worked on Named Entity Recognition (NER), concept normalization, and semantic search, and built scalable data infrastructure including VectorDB indexing pipelines and GraphDB-based systems. I am also experienced in model serving and optimization. I have developed production-grade inference APIs and worked extensively on model lightweighting using tools such as ONNX Runtime and Hugging Face Optimum, focusing on reducing latency and compute cost. I have a strong interest in ML systems performance, particularly model quantization, inference optimization, and ML compilation. While I primarily use Python, I actively explore Rust for its safety and performance guarantees, and apply it in ML-related projects. I contribute to open-source projects in the Rust ML ecosystem: burn (PyTorch-like framework): implemented the Muon optimizer ndarray (NumPy-like library): contributed performance improvements luminal (lightweight DL framework): implemented Conv3D I also build personal projects, including a multi-armed bandit framework (octopus), retrieval models in Rust, and ML tooling. Beyond technical work, I care deeply about code quality and engineering culture, especially code review practices. I have written about the role of AI in code review and its implications for development workflows.