Sangwoo Ha

Summary

ML engineer with 5+ years of experience in medical AI, from model development to regulatory approval and product commercialization. Led a chest X-ray AI project through KFDA clinical trials and published a first-author paper in Nature npj Digital Medicine. Currently responsible for cervical cell analysis AI at Noul, covering the full pipeline from whole-slide image processing and on-device detection to LLM-based diagnostic report generation. Previously developed construction site safety AI at HumanICT.

Experience

ML Researcher — Noul Inc. 2023.08 – Present

Cervical Cell Analysis AI (CER Product) · Yongin, South Korea

Designed and implemented the full preprocessing–detection–classification pipeline for whole-slide microscopy images, from field-of-view segmentation to individual cell classification
Owned on-device detection model lifecycle: training, pruning, quantization, and inference optimization for embedded hardware
Generated pseudo labels via diffusion/SSL-based teacher models; applied teacher-student distillation to improve on-device model accuracy under limited labeled data
Standardized inter-device/slide color variation through stain normalization. Detected annotation errors using t-SNE-based label analysis
Introduced multi-instance learning (MIL) to aggregate cell/patch embeddings into specimen-level diagnosis
Built a RAG pipeline with MedGemma-family LLMs and medical literature for automated diagnostic report draft generation
Independently designed a ResearchOps DAG connecting data processing, training, evaluation, and report generation agents for experiment reproducibility

Software Engineer — HumanICT Co., Ltd. 2022.05 – 2023.08

Alphai PlantVision AI · Seoul, South Korea

Developed real-time detection models for fire, smoke, missing helmets, and vehicle classification on the Alphai PlantVision platform
Compressed and quantized models for resource-constrained CCTV environments while maintaining accuracy
Deployed and validated across multiple construction company sites in production conditions

Software Engineer — DeepNoid Inc. 2020.07 – 2022.05

Chest X-ray AI (DC-XR Series) · Seoul, South Korea

DC-XR-03: Led AI team (part lead) for pneumothorax and consolidation detection. Designed localization-adjusted evaluation methodology requiring lesion-level accuracy. Passed KFDA clinical trials, obtained regulatory approval (Permit No. 21-841), commercialized as product
Published clinical results as first author in Nature npj Digital Medicine
DC-XR-04: Extended detection to 7+ thoracic conditions including lung cancer, atelectasis, cardiomegaly, pulmonary nodules, and pleural effusion

Publication

Lee, S.Y., Ha, S., Jeon, M.G. et al. "Localization-adjusted diagnostic performance and assistance effect of a computer-aided detection system for pneumothorax and consolidation." npj Digital Medicine 5, 107 (2022). doi:10.1038/s41746-022-00658-x — First Author

Skills

Languages Python (primary, 6+ yrs), C/C++, Shell Script, TypeScript, Java ML / DL PyTorch, TensorFlow, ONNX, TensorRT, model pruning/quantization, knowledge distillation Medical AI Whole-slide image analysis, object detection (YOLO, Faster R-CNN), classification, MIL, stain normalization LLM / Agent RAG pipelines, LLM-based report generation, tool-calling, agent orchestration, ResearchOps DAG Data NumPy, Pandas, OpenCV, t-SNE, pseudo labeling, diffusion-based augmentation, label error detection Infra Linux, Docker, Git, CI/CD, on-device deployment, GPU cluster management AI Tools Claude Code, Cursor, Codex CLI, Gemini CLI

Education

M.S. Computer Science — UNIST 2017.03 – 2020.02

Thesis: "Predicting the Location of Injured People in Disaster Zones using Deep Learning"
Proposed gCBAM (gated CBAM) attention and Feature-Highlight Annotation for casualty density estimation. GPA 3.33/4.3

B.S. Electronic Engineering & Computer Engineering (Double Major) — Kyung Hee University 2012.03 – 2016.02

GPA 3.08/4.3

Open Source & Side Projects

oh-my-claudecode — Docs maintainer. Multi-agent orchestration layer for Claude Code (186k+ stars)
oh-my-humanizer — Open-source tool for detecting and removing AI writing patterns. Korean/English support
devswha.xyz — Portfolio site built with Next.js 16, React 19, Tailwind v4
Smart store operation, solo development of SaaS and PT scheduling service. Daily use of Claude Code, Codex, Gemini CLI. Sharing AI coding tips on Threads (@vibe.tip)

Awards & Certifications

Seoul Ideathon Grand Prize (Seoul Mayor's Award), 2016
Kookmin Univ. Startup Camp Award, 2016
Kyung Hee Univ. Design Competition Bronze, 2016
Graduation Thesis Creative Design Grand Prize, 2015
TOEIC 835 (2016)
G-TELP Level 2 Pass (2015)

Languages

Korean Native English Professional working proficiency — conducted research under English-speaking advisor at UNIST; collaborated with international team members