I'm working toward becoming a researcher at the intersection of AI and mechanical engineering, applying data-efficient learning and careful verification to physical-system problems. The work below spans acoustic energy harvesting, a hands-on robotics capstone, optimization, and fault-tolerant control, alongside the tooling I built to keep that work reproducible.
Research at the meeting point of AI and mechanical engineering.
New here? Portfolio Pieces opens like a file browser. Start with Projects for the research, or Live Web Apps if you'd rather click something than read.
Broadband acoustic energy harvesting, where the contribution is steadiness across environments (power variation CoV 21% vs 29% across 11 measured noise environments; advantage grows with resonator quality factor). A Human-in-the-Loop, risk-aware bidding system for a virtual power plant, built with a team at a Korea University × Upstage hackathon (1st place, 23 teams), where I led the system architecture, objective function, and verification/feedback loop. A team-capstone autonomous pole-climbing and coconut-harvesting robot, where I led the control software and power electronics. A decision study on whether a device lab should adopt sequential Bayesian optimization: on three public device/thin-film datasets it reached the random baseline's 40-measurement result in 22, but the lab's own data carries 95% of its variance in measurement noise, so the recommendation was against adopting yet. Asymmetric-geometry rotor fault tolerance. A short-range air-defense interception simulator with a playable in-browser demo, where four optimization methods are compared under proportional-navigation guidance.
Living Paper, a meeting assistant whose MVP is live: it marks the moment you pretended to understand and explains it afterwards from your own lab's documents, with the cloud relaying ciphertext only. Acoustic Harvester, an interactive version of the harvesting study. Divert-Sim, a playable interception simulator. VPP Trade-Guard, a case study of the hackathon system. Engineering Brain, which poses engineering questions and grades the answers. News Literacy Lab, which breaks a news article down by stakeholder. Emotion Quadrant, a Korean-language tool for naming a feeling precisely. 357 Stick Breaker, a misère Nim variant played against a solver. An Optimal Design Visual Guide in Korean, where every chart of the CAE optimization stack is computed live in the browser.
Safeguards designed so no tool grades itself: research change records are extracted from the git diff, regression gates replay real messages through real handlers instead of checking liveness, remote automation passes approval steps with per-thread heartbeats and terminal fallback, and a physics-equation catalogue is graded by evidence (reference, dimensional, numeric) behind a 95% CI gate. Low-quality experiments get dropped, and the engineering-reasoning dataset fed a campus poster separating what transfers into a model from what does not.
Euichan Lee (Leechan), Korea University, Mechanical Engineering and Applied Artificial Intelligence. Undergraduate researcher at the Materials Innovation Lab (Prof. Byungseok Seo). Contact leechan0608@korea.ac.kr.