Résumé

National Cheng Kung University, Taiwan · Sep 2024 — Jun 2027 (expected)

B.S. Purdue Dual Degree Program, Computer Science & Information Engineering · GPA 4.06 / 4.30

Relevant coursework: Linux System and Open Source Software

Co-Founder & CTO, TREKX (ChillOut Travel Planner) · Feb 2026 — Present

Lead all engineering as CTO, owning system architecture and the technical roadmap across a B2B SaaS platform and consumer apps.

Architected a multi-tenant SaaS for travel agencies (Python/FastAPI + Next.js/TypeScript, Dockerized) with an LLM-powered importer that parses unstructured Word itineraries into structured day-by-day plans.

Shipped ChillOut, published on the iOS App Store (React Native/Expo, Express, Python FastAPI microservices with OpenCV/U-2-Net), plus a trilingual travel-personality quiz with OpenAI-streamed analysis and live two-person matching (tbtitest.com).

Research Assistant, National Yang Ming Chiao Tung University (NYCU) · Feb 2026 — Dec 2026

Curated and augmented training datasets for an RL-driven Active Directory honeypot, expanding the PowerShell command corpus and mapping ~1,000 command signatures to MITRE ATT&CK techniques across 14 tactics.

Co-authored research reports (2025–2026) commissioned by the National Chung-Shan Institute of Science and Technology (NCSIST) on active cyber-defense deception techniques, documenting system architecture, experimental methodology and evaluation results.

TSMC SWE Intern · Jul 2026 — Aug 2026

Designed an agentic LLM workflow that auto-generates regression test cases for a legacy VB.NET monorepo (~550k hand-written lines, 1,700+ files), establishing the coverage required for its migration to a containerized (Kubernetes) architecture.

Built a project-agnostic orchestrator that explores the codebase via fuzzy search over abstract syntax trees, self-recovers from failed model calls, and exposes project-specific operations and acceptance criteria as abstract classes.

Adopted by a project spanning six release units, cutting the target delivery time for regression test cases by ~30%.

GMC-Link: Referring Multi-Object Tracking via Global Motion Compensation · Python, PyTorch, OpenCV, Computer Vision

Designed a plug-in module that removes camera ego-motion (homography estimation with RANSAC) and aligns the residual object motion with language expressions via contrastive learning, attaching to existing RMOT trackers without retraining them.

Raised moving-class HOTA by +9.44 points (iKUN 27.70 → 37.14) on Refer-KITTI, and improved pooled HOTA in all four host settings across three RMOT architectures (iKUN, TransRMOT, FlexHook V1/V2).

Autonomous Smart Car Ecosystem · C, Flutter (Dart), Embedded Systems

A Flutter control hub over wireless UART to PIC18F firmware via Bluetooth Serial; Gemini 2.0 Flash parsing unstructured voice input into vehicle protocol commands at over 90% zero-shot accuracy; and a custom driver from the reverse-engineered BLE protocol of a commercial light stick.

Yoban: Real-time Karaoke & Live Streaming Platform · Flutter, Firebase, Azure, Node.js

Automated CI/CD with GitHub Actions covering APK generation and web deployment to GitHub Pages, and a serverless backend on Azure Functions and Node.js securely issuing RTC authentication tokens.

Kungnection: Social Networking Platform for Developers · React, TypeScript, Vite, Java

A type-safe single-page application on React and Vite, with a Java-based remote execution service via the Judge0 API for secure multi-language compilation and real-time browser output.

Frameworks & libraries: React, Next.js, React Native, Flutter, Node.js, FastAPI, PyTorch, OpenCV, Unity 3D

Cloud & tools: Azure, Firebase, Fly.io, Git, Vite, Docker

Specialties: Full-stack web/mobile development, AI/LLM applications, computer vision

18th NCKU Bike Festival — Head of Academic Department · Aug 2024 — Mar 2025

Directed a team of 17 and coordinated within a 113-person student organisation to execute the university’s largest academic exposition, serving over 15,000 attendees.

Achieved a 57% increase in average attendance per seminar by analysing past participation data to restructure topics around student interest.