Yuna Sim

Research internship at Nota AI's Berlin office (Nota GmbH, Europe AI Research), working on Vision-Language-Action (VLA) models for robotic manipulation. - Designed and implemented action decoding modules on a VGGT-based 3D-aware VLA policy, including experiment design, ablation registration, and evaluation on the LIBERO manipulation benchmark - Built tooling for training-scope control and checkpoint provenance tracking, making narrow-scope ablations reproducible and surfacing several silent failure modes -

Data science work at a Korean VPP (Virtual Power Plant) aggregator, spanning electricity market data infrastructure and forecasting. - Built ETL pipelines ingesting KPX (Korea Power Exchange) market data into a Bronze/Silver layered MySQL schema, orchestrated with Airflow - Developed solar generation and SMP (System Marginal Price) forecasting models, with experiment tracking via MLflow - Documented the pipeline and electricity market domain knowledge as team-facing guides

- Completed intensive program covering data analysis with Python, LLM service development, and RAG pipeline engineering - Built "Welfare Compass," a RAG-based chatbot using LangChain, LangGraph, ChromaDB, and GPT API to help citizens navigate 300+ welfare policies - Advanced to finals (Top 20/181 teams) at the Seoul AI Hackathon; recognized for practical applicability

- Conducted research on color patterns in Korean dramas and their influence on emotional engagement using HSV/LAB color analysis and computational methods - Designed user surveys and performed statistical analysis on viewer preferences across original and remake versions - Co-authored peer-reviewed paper published in Journal of the Korean Data Analysis Society (Feb 2025)

- Designed deep learning models for auto white balance using multispectral and multi-view data - Led industry-partnered project on spectral-based image enhancement techniques - Co-authored 2 patents and 2 publications on illumination estimation algorithms - Contributed to broadcast production using AI deep learning technology (MBC)

- Developed YOLOv5-based welding defect detection system achieving 87.5% mAP - Generated synthetic training data using CycleGAN to overcome data scarcity - Co-authored paper presented at IPIU 2024

• Conducted user interviews and competitive analysis to support A-Station platform improvements • Designed e-commerce growth strategy for Evezary

• Thesis: Global illumination estimation using local patch selection • Advisor: Prof. Jewon Kang (ICPL) • Research focus: Computer Vision, Multispectral Imaging, Auto White Balance