Experienced AI Engineer specializing in optimization, LLMs, and computer vision. I believe development is about finding the most appropriate solution—whether elegant code or strategic processes.
My recent work has focused on AI-driven image pre-translation pipelines, where I successfully reduced translation and OCR costs by 96% and 75% respectively while improving pipeline inference speed by 20%.
I thrive in collaborative environments where technical excellence meets practical business outcomes, constantly seeking the optimal path between theoretical possibilities and implementable solutions.
- 🔍 Image Pre-Translation Pipeline: Reduced translation costs by 96%, OCR costs by 75%, improved inference speed by 20%
- 🤖 Korean Small LLM (sLLM): Decreased VRAM usage by 43%, training time by 51%
- 👁️ On-Device Security Solution for Samsung Display: Achieved 97% facial recognition accuracy with 90% less memory
- 🚀 Train/Inference speed optimization for big-models (exllama, Stable Diffusion, vllm, etc)
- 🧠 Finetuning LLMs (full-finetuning, LoRA, QLoRA)
- 📊 Large model Quantization into 4bit (GPTQ, AWQ)
- 📈 Few-data learning using CrossValidation & Stacking Ensemble
- 🤖 Agentic LLM
- 🌟 More Interests sorted by category
- Web Portfolio last updated: 2025.03
Development happens beyond computers. When faced with a problem, I choose the most effective solution—whether it's code or real-world action. This approach guided me in developing the Korean small Language Model (sLLM) and creating the on-device abnormal behavior detection solution for Samsung Display.
- E-mail: mohomin123@gmail.com
- LinkedIn: https://www.linkedin.com/in/hyeongmin-moon-09aaa3164/