Summary
Daehan Kim is an AI engineer based in the Netherlands with nine years of experience building production-grade ML systems spanning TTS, dialog agents, and educational AI. He has led end-to-end model development—from dataset design and post-training of large Korean and roleplay-tuned LLMs to deployment with vllm/Vertex AI and Triton—delivering features that cut hallucinations by ~30% and enable sub-second reasoning latency. At EverAI he architected an in-house long-term memory and multimodal consistency pipeline that links conversation understanding to image generation, and at Neosapience he improved TTS naturalness via DPO and released Korean LLM assets used for semantic search. Comfortable across full-stack ML tooling (PyTorch, Hugging Face, DeepSpeed, FastAPI, BigQuery) and production infra (ArgoCD, k8s), he combines research publications and shipped products, and often optimizes distributed training with profiler-driven tweaks. Notably, he balances rigorous evaluation benchmarks with pragmatic engineering to move experimental models into large-scale user-facing experiences.
9 years of coding experience
5 years of employment as a software developer
Master's degree, Department of Computer and Radio Communications, Master's degree, Department of Computer and Radio Communications at 고려대학교