1. 做物件導向語言(C++)的編寫與設計, 並運用Python作為輔助的資料分析處理語言。 2. 設計與建構高頻量化交易的交易系統與基礎設施 3. 優化目前的交易平台，以達到奈秒等級的進步與改善 4. 設計與實作 Reconcile、Risk control 等等量化交易所需工具 1. Responsible for developing and improving scalable quantitative research frameworks using Python, C++, and other software systems. 2. Designing and implementing a high-frequency trading platform, which includes collecting quotes and trades from and disseminating orders to exchanges around the world 3. Optimizing this platform by using network and systems programming, as well as other advanced techniques to minimize latency 4. Building risk-management and performance-tracking tools

創業家兄弟-松果購物, 資料分析師/Data analyst Intern, Apr 2019 ~ Sep 2019 工作內容：負責SEO 分析顧客搜尋、購買、瀏覽等等大數據（每日有千萬量級的資料量），來強化松果購物電商網站。 我負責的項目有： 顧客網站內的搜尋結果呈現排列優化（分析搜尋行爲後改良Elasticsearch上的SQL語法） 網站本身被搜尋到的SEO（搭配深度學習 增加使用者站外搜尋被引導到松果購物） 搜尋後的推薦系統（運用 PHP , js , python 深度學習） 各種報表的參數蒐集和圖表（運用 Bigquery、Google analyitcs） 使用版本管理器來安全的上傳程式碼（Git） Responsibilities: In charge of SEO, customer searching, purchasing and browsing data analysis (data magnitude of 10 billion per day) in order to improve the performance of PCONE E-commerce platform. What I done: improve merchandise arrangement performance of costumer search in site.(imporve the sql query working on elasticsearch) SEO(With Deep learning skill, increasing PCONE's Google Search Rank) Suggestion system (with PHP , js , python and deep learning skill) useful data analysis and chart making(with Bigquery , Google analyitcs) Use Git push and merge to build a project together 心得：電商領域能充分展現資料分析的作用，例如在我做完站內搜尋結果呈現的優化後，改善後的搜尋結果也為搜尋的收益提高5-10%。這實習的過程中我也學到了許多像是PHP , javascript等等以前不曾接觸過的語言，讓我的努力能夠直接地在網站上被呈現出來。在架構精簡的新創公司，也能更容易去了解一家公司的分工以及成長營運模式。 Review: The power of data analysis can completely be shown in E-commerce.For instance, The revenue cause by customer searching in PCONE website increase for about 5 to 10 percent because of better arrangement. I do learn many skills such as coding with PHP and Javascript for visualizing my effort on website. Moreover , I realize the division of work of a company.Those are valuable experience I learn during my internship.

交通大學, 食物辨識專案/Food Recognition Project, Nov 2018 ~ Mar 2019 專案內容：能夠透過照片來辨識料理的食材組成，是我目前接觸過Deep Learning最有挑戰性的任務。主要的架構最後是選定運用Resnet網路，能夠更容易的去訓練非常深的網絡，過程中碰到許多問題，包括一種食材的多種形狀顏色改變，或者食物的形狀變化。需要透過大量的論文閱讀才能找到解決方法，我在這個專案中學習到不僅僅是電腦圖形識別的問題解決能力，還有比起一般大學生更強的論文搜尋與閱讀能力。 Project Description: One of the most challenging project I have ever done, the food recognition project was centered around trying to figure out the ingredients of dishes through computer vision .The model we use is RESNET, which do better when the model is deep and complicated. I researched and studied a lot of scientific papers related to my project to tackle problems of recognition due to factors of food deformation and the variety of breeds. I learned not only the computer vision deep learning skill, but also how to research and analyze scientific papers in general. [Click the link for more detail:] https://hackmd.io/oFrkP-3-Q6GlPlmU_xYeQA

工研院ITRI, 人工智慧工程師/AI Engineer Intern, Jul 2018 ~ Mar 2019 工作內容：與警政署合作，負責搭建輕量化的車牌偵測與辨識模型，讓模型能在手機上離線做到即時車牌辨識。運用Keras搭建輕量化的CNN模型，並且與台北市停車管理處索取真實狀況的車牌來做訓練，模型訓練完畢後運用蘋果公司的開源API -- OpenML來將Keras轉移到Iphone 手機上運作。 Job Description: Cooperated with Taiwan National Police Agency with building a light license plate recognition system on an Iphone with the CNN Keras model; afterwards, used OpenML to convert Keras code to an Iphone app for further use. 成果：模型準確率在車牌清晰的情況下能有95%以上的正確率，在燈光昏暗、照片模糊等雜訊較多的情形則有較高的出錯率。在結束實習前，ＡＰＰ已成功安裝到手機上測試，效果良好。 Result: The model I built achieve 95% high correct rate.The 5 percent error occured when the photo is dark or blurry. [Click link fo more detail and Github code:] https://hackmd.io/uVq1yifvTtG8Q5MxUK0ceA

交通大學, 人工智慧專題AI project, Jan 2018 ~ Jan 2019 專案內容：透過CNN對於影像處理的強大能力來達到無人駕駛，給予每個畫面對應的操作數據，例如：加速、左右轉、煞車等作為訓練的參數，讓模型模仿人類開車的方式行駛在道路上。 成功在Airsim開源模擬器上，讓車子在道路上行駛平均10分鐘不會有擦撞等事故發生。 Project Description: By strong ability of computer vision CNN model, unmanned car is future trend.We simply use driving parameters such as accelerate ,turning right or left , brake as our training data, expecting the unmanned car can drive itself as human did. The result is great that the car successfully "drove itself" in Airsim Simulator for at least 10 minutes. [Click link for more detail:] https://hackmd.io/lT7_8qSxSuilWA5YJt0iug