cv

Master of Science

University of Southern California

Computer Science (Artificial Intelligence)

Exchange Student

State University of New York at Oswego

Bachelor of Science

Hanyang University

Computer Science & Interior Architecture Design

• Undergraduate Leadership Scholarship
• Hanyang Talent Excellence Award Scholarship

Research Member

AIoT Group at Seoul National University

Worked on Lens, a sensor-control system that optimizes image acquisition under domain shift. Designed evaluation on real-world datasets and drove targeted data collection to expose failure modes and improve robustness.

Undergraduate Researcher

Multimodal AI Lab at Hanyang University

Explored fine-grained visual recognition with multimodal priors (geography/language). Extended a CLIP/Transformer baseline to inject context and measured consistent gains across few-shot settings.

Intern Researcher

COGA Robotics

Worked on indoor person-following: fused LiDAR and RGB-D in ROS, optimized Deep SORT on Jetson, and profiled perception latency to meet real-time constraints.

SNU Ambient AI Competition — 1st Prize

Seoul National University

Built an on-device real-time blurring app (team of 4).

DB GAPS Challenge — 2nd Place

DB Kim Jun Ki Cultural Foundation

Designed automated investment strategy using financial & market data (team of 4).

Lens: Sensor-Aware Image Acquisition under Shift

Prototype sensor control that adapts capture parameters to scene/domain shift to stabilize downstream model quality.

• Evaluation on natural distribution shifts (real-world image sets)
• Data collection protocol to expose failure modes
• Quality tracking with regression-safe comparisons

On-Device Real-Time Blurring (SNU Ambient AI 1st Prize)

Mobile pipeline for privacy-preserving video blurring with tight latency and power budgets.

• Model/runtime trade-off exploration on device
• Deterministic benchmarking harness (latency & quality)
• Team execution from prototype to demo

Few-Shot Fine-Grained Recognition with Context

Injected geographic/language priors into a CLIP-style baseline for fine-grained categories.

• Context feature engineering and conditioning
• Few-shot evaluation with stratified splits
• Ablations + seeded reproducible runs

Indoor Person-Following Stack (ROS + Jetson)

Fused LiDAR and RGB-D to stabilize tracking in cluttered indoor environments.

• ID-switch analysis & latency budget profiling
• Recovery behaviors validated in field tests

OLAP vs. OLTP DBMS Analysis: DuckDB vs. PostgreSQL

System-level comparison of columnar DuckDB and row-store PostgreSQL for analytical vs. transactional patterns.

• Benchmarked scan/buffer/file-format paths with C++ micro-workloads
• Sequential-query performance and trade-off analysis for analytics use cases
• Actionable guidance for pipeline DBMS selection

Classifying Similar Furniture

End-to-end pipeline for visually similar furniture categories with custom image assets.

• Self-created dataset via modeling/rendering; rigorous splits
• TensorFlow baseline → model refinements with measurable gains

Zoning Generation Program for Ideal Floor Plan

Algorithmic zoning assistant for early-stage floor-plan design (Processing).

• Encoded architectural constraints into search/heuristics
• Brought interior-design domain knowledge into computational tooling

“Haenggarae” Social-Value Platform Deployment

Campus-scale rollout of a blockchain-based app incentivizing eco-friendly behaviors.

• Localization and QR/NFC behavior tracking for 10+ activities
• Early user-engagement strategies with industry partner