I build the software that tests self-driving cars in simulation before they reach the road, from the scenario runners to the scheduler that spreads test runs across a GPU cluster. I've also spent years running Linux servers and a department network.
我做自駕車的模擬測試：車子上路前，先丟進各種情境裡跑。從執行情境的程式，到把測試分派到 GPU 叢集的排程系統都寫過；另外也管了好幾年的 Linux 伺服器和系館網路。
I finished my M.S. in Computer Science and Engineering at National Yang Ming Chiao Tung University (NYCU) in August 2026. Through my advisor's collaboration with Taipei Veterans General Hospital, I'm currently building clinical information systems that streamline the care team's workflow. I'm looking for a software engineering role.
2026 年 8 月自陽明交通大學資訊科學與工程研究所畢業。目前透過指導教授與臺北榮民總醫院的合作，協助建置醫療資訊系統，優化醫療團隊的工作流程。正在尋找軟體工程師的工作機會。
Coursework: Deep Learning (A+), Deep Learning Lab (A+), Image Processing (A+), Computer Vision, User Experience and Usability Evaluation.
修課：深度學習（A+）、深度學習實驗（A+）、影像處理（A+）、電腦視覺、使用者經驗與易用性評估。
Autonomous-driving testing and scenario generation; ran the lab's SLURM GPU cluster.
自駕車測試與情境生成研究；維運實驗室的 SLURM GPU 叢集。
Built an internal workflow system from scratch with React, FastAPI, and GitHub Actions CI/CD.
從零寫了一套內部工作流程系統（React、FastAPI、GitHub Actions CI/CD）。
TA for Combinatorial Mathematics (2022) and Network Planning & Management / CCNA (2023). Taught STL and data structures at two camps: at the PCCA winter camp I also set up the online judge, and for the first BPCEI programming camp I wrote problems and test data.
「組合數學」（2022）、「網路規劃管理與實務 CCNA」（2023）課程助教。在 PCCA 冬令營和 BPCEI 第一屆資訊營教 STL 與資料結構；PCCA 的 Online Judge 是我架的，BPCEI 則負責出題和生測資。
Runs the same driving scenarios against different simulators (CARLA, esmini, MetaDrive) to test the planning and control stack of Autoware and other AV systems. I designed the scaling layer: a Rust task queue (Actix, SeaORM, PostgreSQL), a Go scheduler that submits work to SLURM, and a Python executor that orchestrates simulator and AV containers over gRPC.
讓同一組駕駛情境可以在不同模擬器（CARLA、esmini、MetaDrive）上執行，用來測試 Autoware 等自駕系統的 planning 與 control。擴展層是我設計的：Rust 任務佇列（Actix、SeaORM、PostgreSQL）、將工作送進 SLURM 的 Go 排程器，以及透過 gRPC 協調模擬器與自駕系統容器的 Python executor。
Built a 9K-scenario multi-view dataset by composing OpenSCENARIO templates and rendering each from ten viewpoints (ego, drone, infrastructure), then benchmarked Qwen2.5-VL and Cosmos-Reason1 on video-level scenario classification with chain-of-thought and in-context prompting.
組合 OpenSCENARIO 模板產生 9,000 個情境，並從自車、無人機、路側共十個視角算繪成多視角資料集；再以 chain-of-thought 與 in-context learning 評測 Qwen2.5-VL、Cosmos-Reason1 等模型的影片情境分類能力。
An offline-capable iPad web app that replaces a 17-page paper assessment booklet, designed for Apple Pencil input. TypeScript, Vite, and a Node + PostgreSQL backend, deployed with Docker and nginx.
把 17 頁紙本評估表搬上 iPad 的 PWA，支援 Apple Pencil 手寫、可離線使用。TypeScript、Vite，後端 Node + PostgreSQL，以 Docker 與 nginx 部署。
Bridges the esmini OpenSCENARIO player to Autoware, so scenario files can drive Autoware's planning and control in closed loop.
把 OpenSCENARIO 播放器 esmini 接上 Autoware，讓情境檔可以閉迴路驅動 Autoware 的 planning 與 control。
Connected ITRI's in-house driving simulator to Autoware so its planning and control could be tested in closed loop: the simulator speaks ROS 1, Autoware speaks ROS 2, so I bridged ego state, detected and predicted objects with custom messages, and packaged the whole stack as Docker Compose services.
把工研院自研的駕駛模擬器接上 Autoware，讓它的 planning 與 control 能在閉迴路下測試。模擬器走 ROS 1、Autoware 走 ROS 2，我用自訂 message 橋接自車狀態、偵測與預測物件，並把整套環境包成 Docker Compose 服務。
A C++ reimplementation of the nuScenes dataset SDK with Python bindings, for faster indexing and queries on large autonomous-driving datasets.
以 C++ 重寫 nuScenes 資料集 SDK 並提供 Python 介面，加速大型自駕資料集的索引與查詢。
Bootloader, memory allocator, and kernel written from scratch in Rust for a Raspberry Pi 3 (runs on QEMU too).
用 Rust 從零寫給樹莓派 3 的 bootloader、記憶體配置器與 kernel，也能在 QEMU 上執行。