[{"data":1,"prerenderedAt":504},["ShallowReactive",2],{"$fwuenr1OZcDw8TjcC-BzsU1lhSGBFolBhwdGCSmwMXKw":3,"$fE_4--MRHCnRTp2OKqlBfTdkvR4YEdpwy3L9DTivjpko":197,"$fVqO0k1BAadQrWkzSdzKIzB9FEw46o3cTAskVE807S0o":430,"mdc--jtoxnq-key":471},[4,17,30,40,52,62,72,82,90,100,109,119,132,141,150,160,170,178,188],{"id":5,"level":6,"link":7,"publish":8,"reward_type":9,"reward_data":10,"name":11,"intro":14,"image":16},"臺北市資訊局","thanks","https:\u002F\u002Fdoit.gov.taipei",true,"Null",0,{"zh":12,"en":13},"臺北市政府資訊局","Department of Information Technology, Taipei City Government",{"zh":15,"en":15},"臺北市政府資訊局（Department of Information Technology）為臺北市政府資訊業務主管機關，致力於推動市政創新與科技應用，將先進數位科技融入城市治理、公共服務及基礎設施建設，透過各項資訊平台運用，提升民眾良好的市政服務體驗，積極打造更智慧、便捷的臺北城市環境","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1iVTTBCvVuA95ntyn389Gj-nra-dmSRjc",{"id":18,"level":19,"link":20,"publish":8,"reward_type":21,"reward_data":22,"name":23,"intro":26,"image":29},"andes","friend","https:\u002F\u002Fwww.andestech.com\u002Fen\u002F","連續贊助",3,{"zh":24,"en":25},"晶心科技股份有限公司","Andes Technology Corporation",{"zh":27,"en":28},"晶心科技股份有限公司于2005年成立於新竹科學園區，2017年於臺灣證交所上市 ([TWSE: 6533](https:\u002F\u002Ffinance.yahoo.com\u002Fquote\u002F6533.TW?p=6533.TW&ncid=stockrec); [SIN: US03420C2089](https:\u002F\u002Fwww.bourse.lu\u002Fsecurity\u002FUS03420C2089\u002F342557); [ISIN: US03420C1099](https:\u002F\u002Fwww.bourse.lu\u002Fsecurity\u002FUS03420C2089\u002F342557))。晶心是RISC-V國際協會的創始首席會員，也是第一家推出商用RISC-V向量處理器的主流CPU供應商。為滿足當今電子設備的嚴格要求，晶心提供可配置性高的32\u002F64位元高效能CPU核心，包含DSP、FPU、Vector、超純量  (Superscalar)、亂序執行  (Out-of-Order)、多核心及車用系列，可應用於各式ＳｏＣ與應用場景。晶心並提供功能齊全的整合開發環境和全面的軟\u002F硬體解決方案，可幫助客戶在短時間內創新其SoC設計。截至2025年底，Andes-Embedded™ SoC累計出貨量已超過200億顆。欲瞭解更多資訊，請訪問  [https:\u002F\u002Fwww.andestech.com](https:\u002F\u002Fwww.andestech.com\u002F)。請立即透過[LinkedIn](https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002F13688177\u002F)、[Facebook](https:\u002F\u002Fwww.facebook.com\u002FAndesTechTW?locale=zh_TW)、[X(原 Twitter)](https:\u002F\u002Ftwitter.com\u002FAndes_Tech)、[YouTube](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FAndesTechnology\u002F) 以及[Bilibili](https:\u002F\u002Fspace.bilibili.com\u002F335295020)追蹤晶心最新消息。","As a Founding Premier member of RISC-V International and a leader in commercial CPU IP, Andes Technology ([TWSE: 6533](https:\u002F\u002Ffinance.yahoo.com\u002Fquote\u002F6533.TW?p=6533.TW&ncid=stockrec); [SIN: US03420C2089](https:\u002F\u002Fwww.bourse.lu\u002Fsecurity\u002FUS03420C2089\u002F342557); [ISIN: US03420C1099](https:\u002F\u002Fwww.bourse.lu\u002Fsecurity\u002FUS03420C2089\u002F342557)) is driving the global adoption of RISC-V. Andes’ extensive RISC-V Processor IP portfolio spans from ultra-efficient 32-bit CPUs to high-performance 64-bit Out-of-Order multiprocessor coherent clusters. With advanced vector processing, DSP capabilities, the powerful Andes Automated Custom Extension (ACE) framework, end-to-end AI hardware\u002Fsoftware stack, ISO 26262 certification with full compliance, and a robust software ecosystem, Andes unlocks the full potential of RISC-V, empowering customers to accelerate innovation across AI, automotive, communications, consumer electronics, data centers, and mobile devices. Over 20 billion Andes-powered SoCs are driving innovations globally. Discover more at [www.andestech.com](http:\u002F\u002Fwww.andestech.com\u002F) and connect with Andes on [LinkedIn](https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002F13688177\u002Fadmin\u002Fdashboard\u002F), [X (formerly Twitter)](https:\u002F\u002Fx.com\u002FAndes_Tech) , [YouTube](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FAndesTechnology) and [Bilibili](https:\u002F\u002Fspace.bilibili.com\u002F335295020).","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1SXvUlmHuSXDchd6HoejXMvHiDebVoVgL",{"id":31,"level":6,"link":32,"publish":8,"reward_type":9,"reward_data":10,"name":33,"intro":36,"image":39},"sifive","https:\u002F\u002Fwww.sifive.com",{"zh":34,"en":35},"美商賽發馥股份有限公司臺灣分公司","SiFive",{"zh":37,"en":38},"身為 RISC-V 的發明者與領導廠商，SiFive 正在改變未來運算的典範，將 RISC-V 的無限潛力引領至世上最高效能與資料密集應用中。SiFive 所建構無與倫比的運算平台能夠在晶片設計的每個細分市場成功協助全世界眾多科技領導廠商，從創新發明、優化至完美、並推出最先進的晶片設計解決方案，涵蓋人工智慧、機器學習、車用電子、資料中心、行動運算與消費性電子等應用領域。SiFive 與您同行，創造 RISC-V 的無限未來。欲知更詳細信息，請洽 www.sifive.com","As the pioneers who introduced RISC-V to the world, SiFive is transforming the future of computing by bringing the power and flexibility of RISC-V to the world. SiFive’s market-leading IP provides the blueprint for high-performance, customizable, and energy-efficient processor cores across the entire computing spectrum, from the intelligent edge to the most advanced AI data centers. For more information, visit www.sifive.com","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=17TquX8coFopolRFi2W9ZbDUzgkT5YF0X",{"id":41,"level":42,"link":43,"publish":8,"reward_type":21,"reward_data":44,"name":45,"intro":48,"image":51},"IMA","gold","https:\u002F\u002Fwww.ima.org.tw\u002F",2,{"zh":46,"en":47},"中華民國資訊經理人協會","Information Management Association",{"zh":49,"en":50},"IMA資訊經理人協會自1982年成立至今已逾四十年，是台灣歷史最悠久的IT專業團體之一。自2022年由台灣大哥大資訊長蔡祈岩接任理事長後，致力於串聯IT產業、產業IT、政府、學術及開源社群資源，推動台灣成為全球最能發揮IT與AI專業實力的國家。\n\n近年協會重點推動工作如下：\n一、IT Matters Awards—打造IT雇主品牌獎：由數位發展部指導、IMA主辦的 IT Matters Awards，是全台首創以IT人才、技術與應用為核心的年度獎項，表揚重視IT人才、推動數位轉型並對社會產生正向影響的企業與個人。2026年將持續擴大辦理並與《天下雜誌》共同舉辦，透過評選標竿企業、人才與創新專案，分享最佳實務經驗，促進企業與人才交流，推動台灣IT人才質與量的雙向成長。\n二、執行政府數位發展相關計畫：整合活動推廣、產業交流、Demo Day與多元行銷策略，今年度更協助軟體開源推廣機制，包含開源上架指引優化等，擴大產業參與及政策影響力。\n三、Taiwan Tongues台灣通用語料集計畫：建構台灣通用語料共享計畫，致力建立涵蓋台灣華語、台語、客語與原住民族語的開放語料庫，讓全球語言模型能理解台灣的語言與文化。\n四、台灣開源連線計畫：舉辦專家座談、開源專案展示與國際交流，攜手台灣開源社群推動技術交流與創新合作，打造「台灣開源隊」，提升我國在全球開源生態系中的角色。\n\nIMA將持續以專業行動，深化產業鏈結與生態系建構，從單一活動推動提升為跨產業協作平台，為台灣在全球數位與AI浪潮中爭取更關鍵的位置。","Information Management Association (IMA), Taiwan\nFounded in 1982, IMA is one of Taiwan's longest-established IT professional organizations, dedicated to strengthening Taiwan's digital competitiveness.\n\nSince 2022, the association has been chaired by Rock Tsai, CIO of Taiwan Mobile.\nOur members:\n．Industrial Enterprises: AMD, Phison, Taiwan Mobile, Fubon Bank, etc.\n．IT Solution Providers: Top 10 SI companies such as Systex Group, WITS, Syscom Group\n\nIn 2026, IMA focuses on four key initiatives:\n\n1. IT Matters Awards: Recognizing outstanding IT employers, technology talent, digital transformation achievements, and innovative projects, co-organized with CommonWealth Magazine to share best practices and foster collaboration.  \n2. Government Digital Development Programs: Supporting policy implementation through industry events, networking activities, Demo Days, and multi-channel communications.\n3. Taiwan Tongues: Building an open corpus covering Taiwanese Mandarin, Taiwanese Hokkien, Hakka, and Indigenous languages to help global language models better understand Taiwan.\n4. Taiwan Open Source Initiative:  Promoting technical collaboration, open-source project showcases, and international exchange; building “Team Taiwan” in the global open-source ecosystem.  \n\n\nIMA remains committed to advancing IT value, fostering cross-sector collaboration, and positioning Taiwan at the forefront of the global digital and AI landscape.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1USFHc7t-l4bctCkZUXQ_f21Fng1ZO1lR",{"id":53,"level":54,"link":55,"publish":8,"reward_type":9,"reward_data":10,"name":56,"intro":58,"image":61},"KKTIX","bronze","https:\u002F\u002Fkktix.com\u002F",{"zh":57,"en":53},"華娛網路娛樂股份有限公司",{"zh":59,"en":60},"KKTIX作為台港兩地領先的活動售票、報名與線上直播平台，致力於透過穩定\n可靠的技術服務，協助主辦單位打造更好的活動體驗。從技術研討會、開源社\n群聚會、教育課程到大型論壇、展演活動與線上直播，KKTIX支援各種規模的\n活動需求。\n自2010年成立以來，KKTIX已累積超過700萬名會員，每年服務超過75,000場\n活動。平台提供報名管理、售票、金流、會員經營、數據分析等全方位方案，\n讓主辦單位能專注於內容與社群經營！\nKKTIX長期支持技術與開源社群發展，陪伴許多開發者社群、使用者社群及非\n營利組織舉辦活動。我們相信，每一場活動都是知識交流與社群連結的起點，\n而好的工具能讓更多人專注於分享、學習與創造價值。","As Taiwan and Hong Kong’s leading platform for event ticketing, registration,\nand live streaming, KKTIX is dedicated to empowering organizers with reliable\ntechnology and seamless event experiences. From technical conferences,\nopen-source community meetups, and educational programs to large-scale\nforums, live performances, and online events, KKTIX supports events of all\nsizes and formats.\n\nSince its founding in 2010, KKTIX has amassed more than 7 million registered\nmembers and proudly supports over 75,000 events annually. The platform\nprovides comprehensive solutions—including event registration, ticketing,\npayment processing, membership management, and data analytics—enabling\norganizers to focus on curating impactful content and fostering vibrant\ncommunities.\n\nKKTIX has a long-standing commitment to driving the growth of technology\nand open-source communities, partnering with developer ecosystems, user\ngroups, and non-profit organizations to bring people together. We believe that\nevery event serves as a catalyst for knowledge sharing and community\n\nbuilding, and that the right tools empower people to focus on learning,\ncollaboration, and creating lasting value.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1rWagmNMioYe6vRW0144D939XqHD-ZhI-",{"id":63,"level":54,"link":64,"publish":8,"reward_type":9,"reward_data":10,"name":65,"intro":68,"image":71},"國網中心","https:\u002F\u002Fwww.nchc.org.tw",{"zh":66,"en":67},"財團法人國家實驗研究院國家高速網路與計算中心","National Center for High-performance Computing",{"zh":69,"en":70},"財團法人國家實驗研究院國家高速網路與計算中心（簡稱國網中心），於1991年成立，致力扎根國內高速計算技術，提供世界級的高速計算與學研網路設施，為台灣的科技能量奠基。我們秉持著「驅動轉型，為更美好的未來而努力」的信念，積極推動高速計算的技術與應用發展，引領數位轉型並促進智慧生活的改變。","National Center for High-performance Computing (NCHC), founded in 1991,is committed to strengthening national high-performance computing (HPC)technology, providing world-class supercomputing and academic researchnetwork infrastructure, and laying foundation for the development oftechnology in Taiwan. NCHC, in the spirit of \"driving transformation fora better future with HPC\", is dedicated to promoting HPC technology andapplications development, leading digital transformation in industriesand smart living.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1zegCBM3yK8M049xewjkc0IRtE2UqIUdU",{"id":73,"level":42,"link":74,"publish":8,"reward_type":21,"reward_data":44,"name":75,"intro":78,"image":81},"國泰金控","https:\u002F\u002Fwww.cathayholdings.com\u002Fholdings\u002Fbrand\u002Ffintech",{"zh":76,"en":77},"國泰金融控股公司","Cathay Financial Holdings",{"zh":79,"en":80},"國泰金控致力成為「以金融為核心的科技公司」，透過數位、數據與技術，積極研發並導入國際前瞻技術，打造創新平台與產品服務。\n整合集團資源、強化一站式數位金融體驗，國泰金控發揮持續創新與技術領先優勢，樹立技術與開發環境典範。同時，積極佈局海外並接軌國際，深耕大中華與東南亞市場，注入數據驅動文化，以高效協作的矩陣式組織，持續朝「亞太地區最佳金融機構」願景邁進，共創更好的未來金融。","Cathay Financial Holdings (FHC) is committed to being a technology company with finance at its core. We actively develop and adopt international innovations to build cutting-edge platforms and services through digitalization, data, and technology.\nBy integrating group-level resources, Cathay FHC leverages its strengths in continuous innovation, intelligent applications, and technological leadership to set a benchmark for development environments. At the same time, we continue expanding across the Asia-Pacific, aligning with global trends to become the leading financial institution in the region. Driven by a data-driven culture and a highly collaborative matrix organization, Cathay FHC is dedicated to shaping the future of finance.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1JN_uNRjQ6-9I7z_azEjhahMtTQqDk3SC",{"id":83,"level":54,"link":84,"publish":8,"reward_type":9,"reward_data":10,"name":85,"intro":86,"image":89},"ONLYOFFICE","https:\u002F\u002Fwww.onlyoffice.com",{"zh":83,"en":83},{"zh":87,"en":88},"ONLYOFFICE 是一個國際性的開源專案，由領先的 IT 公司 Ascensio System SIA 開發。目前在中國、新加坡、英國等 7 個國家設有分公司，為全球企業與個人使用者提供高效率的文件處理與協作解決方案，深受網易、百威中國、中信集團、南京大學、中國知網等眾多企業與教育機構的青睞。\n\nONLYOFFICE Docs 是一套功能完整的線上辦公套件，整合了文件編輯、試算表、簡報、可填寫表單以及 PDF 編輯器，並與 Microsoft Office 格式具備高度相容性。此外，還提供數百種格式與樣式設定工具，以及多元的協作功能。","ONLYOFFICE, an open-source office software project, focuses on advanced and secure office solutions. With over 15 million users worldwide, it is recognized for its innovation in the online office domain. The ONLYOFFICE ecosystem includes collaborative applications such as online editors for text documents, spreadsheets, presentations, forms, and PDFs, along with a room-based collaborative platform. \nAs an international company, ONLYOFFICE has employees and contributors across the globe, with offices located in Singapore, Dallas, Shanghai, Riga, London, Belgrade, Yerevan, and Tashkent.\n","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1uilDLdToDeqOJYEGoljpbw2l9YOYDKvs",{"id":91,"level":54,"link":92,"publish":8,"reward_type":9,"reward_data":10,"name":93,"intro":96,"image":99},"QNAP","https:\u002F\u002Fwww.qnap.com\u002Fzh-tw",{"zh":94,"en":95},"QNAP Systems, Inc. 威聯通科技","QNAP Systems, Inc.",{"zh":97,"en":98}," QNAP 命名源自於高品質網路設備製造商（Quality Network Appliance Provider），我們致力研發軟體應用，匠心優化硬體設計，並設立自有生產線，以提供全面而先進的科技解決方案。QNAP 專注於儲存、網通及智慧視訊產品創新，並持續拓展人工智慧應用，推動 Edge AI 邊緣優先、雲端協作的整合模式，協助客戶在不同場域中即時分析、決策與協同運作，充分發揮數據價值。\n\nQNAP 亦是一個全面性的資安品牌，我們將資料保護、防勒索、不可變儲存與多層資安機制融入解決方案中，確保企業營運不中斷並維持長期資料完整性。在 QNAP 的企業藍圖中，NAS 早已突破儲存裝置的框架，更是驅動人工智慧、邊緣運算與資安防護的重要平台，為全球用戶創造更大優勢與價值。","QNAP (Quality Network Appliance Provider) is a global leader in software development, hardware design, and in-house manufacturing, delivering solutions that help businesses and individuals securely store, connect, and innovate. We drive breakthroughs in storage, networking, and smart video, and advance an Edge AI strategy that integrates edge-first deployment with cloud collaboration—enabling real-time data analysis, decision-making, and collaboration across diverse environments.\n\nQNAP is also a comprehensive cybersecurity brand, embedding data protection, anti-ransomware resilience, immutable storage, and multi-layered security into our solutions. For us, NAS has evolved beyond storage—turning NAS into the platform that drives AI, edge intelligence, and trusted data protection worldwide.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1ZBMNO4GxVf6McP1VBEbkL7qs2xdLHBSm",{"id":101,"level":19,"link":102,"publish":8,"reward_type":9,"reward_data":10,"name":103,"intro":105,"image":108},"Appier","https:\u002F\u002Fwww.appier.com\u002Fzh-tw\u002F",{"zh":104,"en":101},"沛星互動科技股份有限公司",{"zh":106,"en":107},"Appier 是一家 AI 原生的 Agentic AI 即服務（AaaS）公司，透過最先進的廣告科技（AdTech）與行銷科技（MarTech）解決方案，協助企業制定更明確的商業決策。創立於 2012 年，Appier 秉持 Making AI Easy by Making Software Intelligent 的願景，致力透過旗下由 Agentic AI 驅動的的廣告雲（Ad Cloud）、個人化雲（Personalization Cloud) 及數據雲（Data Cloud）解決方案，賦予企業自主、自適應、即時決策的能力，將 AI 轉化為可衡量的投資報酬（ROI）。Appier 目前在亞太、美國與 EMEA 地區設有 17 個據點，並於東京證券交易所 Prime 板上市。欲了解更多資訊請參閱 Appier | Empowering Businesses to Turn AI into ROI 。","Appier (TSE: 4180) is an AI-native Agentic AI as a Service (AaaS) company that empowers businesses to create value through cutting-edge AdTech and MarTech solutions. Founded in 2012 with the vision of “Making AI Easy by Making Software Intelligent,” Appier helps businesses turn AI into ROI through its Ad Cloud, Personalization Cloud, and Data Cloud—each powered by Agentic AI that enables autonomous, adaptive, and real-time decision-making. Today, Appier operates 17 offices across APAC, the US, and EMEA, and is listed on the Tokyo Stock Exchange. Learn more at Appier | Empowering Businesses to Turn AI into ROI .","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1HXbbA7QkJWwkXeHPE0nTfYRJcfKI0E4z",{"id":110,"level":42,"link":111,"publish":8,"reward_type":21,"reward_data":112,"name":113,"intro":115,"image":118},"玉山銀行","https:\u002F\u002Fwww.esunbank.com\u002Fzh-tw\u002Fpersonal",5,{"zh":110,"en":114},"E.SUN Bank",{"zh":116,"en":117},"玉山銀行成立於1992年，取名自台灣最高的山，以顧客體驗為核心的理念，致 力於構建科技與綠色雙軸轉型的的金融發展策略。作為數位金融的領導品牌， 玉山銀行擁有超過1300位科技人才，負責整體的數位發展、智能應用、資訊研 發以及資安管理。為臺灣首家銀行將人工智慧深入應用於各項業務，也是首家 銀行以開放的雲端原生技術、微服務架構自建核心系統。近年來玉山銀行展現 卓越的綜合績效，榮獲第十二屆菁業獎共8項大獎，累計歷屆獲奬總數持續領先 金融同業。從台灣到亞洲，玉山以靈活的策略和高效的執行力穩健發展，為顧 客創造持續的價值，邁向永續經營的未來。","E.SUN Bank, established in 1992 and named after Taiwan’s highest mountain, is dedicated to customer-centric values and committed to developing financial strategies centered around technology and ESG. With more than 1,300 technology professionals forming a technology team responsible for overall digital development, AI applications, IT research, and information security management, we are the first bank in Taiwan to deeply integrate AI into various businesses and the first bank to build our core system using cloud-native technology and a microservices architecture. In recent years, E.SUN Bank has demonstrated outstanding comprehensive performance, winning a total of 8 major awards at the 12th Elite Awards for Taiwan Banking Excellence, which maintain our industry-leading record in total awards received. From Taiwan to Asia, E.SUN Bank steadily grows with flexible strategies and efficient execution, creating sustained value for our customers and moving toward a future of sustainability.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1Wu648Tl50fARXtXGK2qkgPfXWhc4-hiT",{"id":120,"level":121,"link":122,"publish":8,"reward_type":123,"reward_data":124,"name":125,"intro":128,"image":131},"慧邦科技","silver","https:\u002F\u002Fwww.gamesofa.com","累計贊助",16,{"zh":126,"en":127},"慧邦科技股份有限公司  Gamesofa Inc.","Gamesofa Inc.",{"zh":129,"en":130},"慧邦科技Gamesofa成立於2005年，透過優異的網路核心技術，自製研發並行銷營運30餘款web, iOS, Android多人遊戲，是台灣首屈一指的遊戲開發公司。\n\nGamesofa以「5分鐘．想樂最輕鬆」為核心理念，成功打造台灣第一大網路麻將遊戲【神來也麻將】並在印尼、越南、星馬、美加推出德州撲克與當地牌類\n及模擬休閒遊戲【貓咪造咖】，全球累積註冊用戶數超過7,500萬，每日活躍用戶超過200萬人。Gamesofa除了在遊戲開發力求創新，更擅長以資料分析\n驅動營運決策，讓產品開發、營運、行銷團隊皆能透過數據分析工具與預測模型，快速掌握遊戲狀況，擬定策略，是一間充滿熱情與效率的企業。","Founded in 2005, Gamesofa is a leading Taiwanese game developer powered by top-tier core web technologies. We’ve self-developed and operated over 30 multiplayer games across web, iOS, and Android platforms.\n\nLiving by our motto, \"Fun in Five Minutes,\" we successfully launched Taiwan’s #1 online Mahjong game, GodGame Mahjong. We're also capturing hearts across Indonesia, Singapore, Malaysia, and North America with Texas Hold’em, local card games, and our casual simulation hit, Catfe World. With over 75 million registered users worldwide and 2 million+ daily active users, our community keeps growing!\n\nWe don't just innovate; we run on data. By leveraging advanced data analytics and predictive models, our development, operations, and marketing teams can instantly read the game environment and pivot strategies. At Gamesofa, we combine passion with high-efficiency execution.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1Tw5jqRB3RNLQ6ey9dklsM0fuAWzVXW8o",{"id":133,"level":54,"link":134,"publish":8,"reward_type":21,"reward_data":124,"name":135,"intro":137,"image":140},"祐生研究基金會","https:\u002F\u002Fwww.archilife.org\u002F",{"zh":133,"en":136},"The Archilife Research Foundation",{"zh":138,"en":139},"祐生研究基金會自 1987 年推動建築與環境永續發展之研究開始，推動長期的知識累積與人才培養。除了過去建築、環境、生態、健康與資訊社會與文化等面向之碩士論文獎助主題外，本會目前定期舉辦的聯誼會及讀書會，強化祐生成員的知識密度。 其後，於 2011 年起，本會持續針對國內開源活動進行贊助，也針對國內獨立遊戲開發者，推動知識與資訊分享聚會，期能幫助國內相關社群之發展。","Since 1987, the Archilife Research Foundation started long-term researches in architecture, sustainability, and go on to other areas of knowledge integration. Efforts including previously giving scholarships in aforementioned areas, and currently holding reading group meetings and communion events for knowledge keepers to better our understandings of the world. Starting from 2011, we also have been sponsoring various open source related events, holding meetups and game jams for local game developers’ groups, in the hope of supporting the growth of relevant communities in these fields.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1Jhy0M0l7kqfA-auPZwPHmrwxtuiBsrjB",{"id":142,"level":143,"link":144,"publish":8,"reward_type":9,"reward_data":10,"name":145,"intro":147,"image":149},"Canonical","diamond","https:\u002F\u002Fcanonical.com\u002F",{"zh":142,"en":146},"Canonical ",{"zh":148,"en":148},"Helping innovators build the future since 2004.\nCanonical’s story is closely linked to the communities that shape us, including the global community of open source contributors and enthusiasts who are part of our team.\n\nSince 2004, we have nurtured a community around Ubuntu, a Linux distribution that leveled the playing field by making software accessible to everyone - whether you are a developer in Calcutta, an end user in Cape Town or a sysadmin in California. Today, we continue to build and grow communities around us, as we help make open source more secure, reliable and manageable at scale.\n\nTogether, we enable organizations and end users everywhere to innovate with confidence on the largest supported collection of open source software.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=100QqFUM1e5FQoSqOnYtwmfAFBU4uJGNF",{"id":151,"level":42,"link":152,"publish":8,"reward_type":123,"reward_data":153,"name":154,"intro":156,"image":159},"Oracle（MySQL）","https:\u002F\u002Fwww.mysql.com\u002F",12,{"zh":155,"en":155},"MySQL",{"zh":157,"en":158},"MySQL 是世界上最受歡迎的開源資料庫。 它為 Facebook、Booking、Uber 和 Tesla 等最具創新性的公司都用它當主要的資料庫。 它可以在本地使用，也可以作為 Oracle 雲端上的 MySQL 資料庫服務使用。 對於開發人員來說，它是一個易於使用、可靠且高效能的資料庫，支援 NoSQL 和 SQL。許多 SaaS、電子商務、金融、電信和Fortune 1000 強公司都依靠 MySQL 企業版的高級安全功能來保護資訊隱私、防止資料外洩，並協助滿足 GDPR、PCI、HIPAA 等監管要求。 MySQL 高可用性具有內建的 HA能力，完全整合到 MySQL 伺服器中，可實現 99.99% 的正常運作時間\n","MySQL is the World’s Most Popular Open Source Database. It powers the most innovative companies including Facebook, Booking, Uber, and Tesla. It is available on-premises and as the MySQL Database Service on Oracle Cloud. For developers, it’s an easy to use, reliable and high-performance database with NoSQL and SQL support. SaaS, ecommerce, financial, telecom, and Fortune 1000 companies rely on the MySQL Enterprise edition advanced security features to protect the privacy of information, prevent data breaches and help meet regulatory requirements such as GDPR, PCI, HIPAA. MySQL High Availability has native HA, fully integrated into the MySQL Server for 99.99% uptime. ","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1gK-ZMKBj5g59qekvVWutwAq7s2v3MmGT",{"id":161,"level":42,"link":162,"publish":8,"reward_type":123,"reward_data":163,"name":164,"intro":166,"image":169},"Berry AI","https:\u002F\u002Fberry-ai.com",6,{"zh":165,"en":161},"華捷智能股份有限公司",{"zh":167,"en":168},"Berry AI 運用電腦視覺與 AI 技術，協助美國速食（QSR）業者分析得來速服務流程、改善營運效率。我們的產品在全美門市即時運作，從攝影機影像中偵測車輛動線、計時服務流程，並產出營運洞察，工程團隊每天面對的是真實世界、大規模的 Vision AI 挑戰。\nBerry AI 的客戶包含多間全球前十大連鎖速食品牌。繼 2025 年與 Zaxby's 簽署近千家門市的全品牌導入合約後，2026 年我們再與美國知名品牌 Culver's 展開全品牌部署，業務持續快速成長。Berry 亦獲得台灣上市公司飛捷科技的投資（全球前三大 POS 製造商之一），擁有穩定的資源與客戶基礎。\n我們的團隊位於台灣，成員來自海內外知名學府與大型科技公司，工程師能直接參與美國市場產品的核心開發。隨著業務成長，我們也在持續徵才，歡迎造訪 berry-ai.com 了解更多。","Berry AI leverages computer vision and AI to help U.S. quick-service restaurant (QSR) operators analyze drive-thru workflows and improve operational efficiency. Our products run in real time at restaurants across the U.S.—detecting vehicle journeys from camera feeds, timing service stages, and generating operational insights. For our engineers, that means working on real-world, large-scale Vision AI challenges every day.\nBerry AI serves several of the world's top ten fast-food chains. Following our 2025 full-brand agreement with Zaxby's covering nearly 1,000 locations, we kicked off another full-brand deployment with Culver's in 2026, and our business continues to grow rapidly. We are backed by Flytech Technology, a publicly listed Taiwanese company and one of the world's top three POS manufacturers.\nOur team is based in Taiwan, with members from top universities and major tech companies worldwide—engineers here directly contribute to the core of products deployed in the U.S. market. As we grow, we're also hiring; visit berry-ai.com to learn more.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=19MMkcQMO3m6qMBD8KvHSEVIeFdY44WHU",{"id":171,"level":42,"link":172,"publish":8,"reward_type":9,"reward_data":10,"name":173,"intro":174,"image":177},"Nitra"," ",{"zh":171,"en":171},{"zh":175,"en":176}," \"The financial home for independent medicine.\" Nitra 是專為美國醫療院所打造的 AI-native 財務平台，讓每間獨立診所都能少花時間處理繁瑣行政、多花時間照顧病患。\n\n\nNitra 整合財務管理、醫療採購與病患行政系統，建立能自動化支付、採購、庫存管理、預約排程、保險資格驗證與病患溝通等核心流程。成立至今，Nitra 已完成 1.87 億美元（約新台幣 59.8 億元）募資，年化交易額突破 10 億美元，是美國醫療科技領域成長最快的新創之一。\n\n\nNitra 從創立初期便將產品研發重心設於台灣，目前 Nitra Taiwan 主導公司絕大部分的產品開發工作，團隊規模在過去半年成長超過一倍，成員涵蓋軟體工程師、設計師與產品經理，並持續快速擴編中。","\"The financial home for independent medicine.\" Nitra is the AI-native financial platform built for U.S. healthcare specialty practices — helping every independent practice spend less time on paperwork and more time on patients.\n\n\nThe platform brings together financial management, a medical supplies marketplace, and patient administration — automating payments, procurement, inventory management, scheduling, insurance eligibility verification, and patient communication. Nitra has raised $187M in funding and surpassed $1B in processing volume, making it one of the fastest-growing healthcare fintech startups in the U.S.\n\n\nFrom its earliest days, Nitra chose to build its core R&D team in Taiwan. Today, Nitra Taiwan leads the majority of the company's product development, with the local team more than doubling in size over the past six months — spanning software engineers, designers, and product managers — and continuing to expand.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1SPHLtEBEX9ap_snb8Rm2SwjIVGlR52gq",{"id":179,"level":54,"link":180,"publish":8,"reward_type":9,"reward_data":10,"name":181,"intro":184,"image":187},"源鋼技術顧問有限公司","https:\u002F\u002Fsunsquare.tech\u002F",{"zh":182,"en":183},"源鋼技術顧問有限公司 SUN SQUARE Co., Ltd","SUN SQUARE Co., Ltd",{"zh":185,"en":186},"源鋼技術顧問有限公司 Sun Square Co., Ltd. 專注於資安、技術顧問與產業標準導入服務，協助企業強化產品安全、系統安全、軟體供應鏈安全與合規實作能力。\n\n我們長期關注開源治理、SBOM、開源供應鏈安全、工控系統資安與產業標準實作，也持續參與並貢獻開源社群與相關技術實務。我們相信開放技術與工程實務是建立可信賴數位環境的重要基礎。\n\n源鋼歡迎對開源軟體、安全工程、SBOM、供應鏈安全與工控資安有興趣的開發者、研究者與技術夥伴，一起交流與合作。","Sun Square Co., Ltd. provides cybersecurity, technical consulting, and industry standard implementation services to help organizations strengthen product security, system security, software supply chain security, and practical compliance capabilities.\n\nWe have long followed open source governance, SBOM, open source supply chain security, industrial cybersecurity, and practical implementation of industry standards. We also continue to participate in and contribute to open source communities and related engineering practices. We believe open technologies and engineering practices are important foundations for building a trustworthy digital environment.\n\nSun Square welcomes developers, researchers, and technology partners interested in open source, security engineering, SBOM, supply chain security, and industrial cybersecurity to connect, exchange ideas, and collaborate with us.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1ozuvt4cNMXQCQqD5VY4EyMdQSbwVuDAL",{"id":189,"level":6,"link":190,"publish":8,"reward_type":9,"reward_data":10,"name":191,"intro":193,"image":196},"Rozeta","https:\u002F\u002Frozeta.app",{"zh":192,"en":192},"Rozeta AI",{"zh":194,"en":195},"Rozeta AI 是專為全球化溝通打造的即時翻譯平台，提供精準的語音轉文字、雙向翻譯與即時字幕，協助不同語言的參與者同步理解會議內容，並使用自己最熟悉的語言自在交流。\n不同於一般逐字翻譯工具，Rozeta AI 能根據會議主題、產業背景及企業自訂詞彙掌握溝通情境，提升專業術語、產品名稱與特定用語的翻譯準確度。無論是跨國團隊協作、客戶會議、教育訓練、線上研討會、國際論壇或大型活動，都能提供流暢且一致的多語溝通體驗。\n使用者透過電腦、平板或手機瀏覽器即可加入，無需下載或安裝 App，便能即時查看共享逐字稿與翻譯內容。平台亦支援簡報浮動字幕、講者辨識、會議逐字稿與 AI 摘要，協助團隊從即時溝通延伸至有系統的會後整理與追蹤。\nRozeta AI 提供 40 種語言的低延遲即時翻譯，協助企業跨越語言隔閡，提升國際協作效率，讓每位參與者都能充分理解、清楚表達並真正投入對話。","Rozeta AI is a real-time translation platform designed for global communication. It delivers accurate speech-to-text transcription, bidirectional translation, and live captions, enabling participants who speak different languages to follow meetings in real time and communicate confidently in the language they know best.\n\nUnlike conventional word-for-word translation tools, Rozeta AI understands the context of each conversation by incorporating meeting topics, industry knowledge, and organization-specific terminology. This improves the accuracy of technical terms, product names, and specialized language. Whether used for global team collaboration, client meetings, training sessions, webinars, international forums, or large-scale events, Rozeta AI provides a seamless and consistent multilingual communication experience.\n\nParticipants can join through a web browser on a computer, tablet, or mobile device without downloading or installing an app. They can instantly access shared live transcripts and translations, while additional features such as floating captions for presentations, speaker identification, meeting transcripts, and AI-generated summaries support structured post-meeting review and follow-up.\n\nWith low-latency, real-time translation across 40 languages, Rozeta AI helps organizations overcome language barriers, improve global collaboration, and empower every participant to understand fully, communicate clearly, and engage meaningfully in every conversation.","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1EgrBD5uRahGcIBjIl7CC0off_2uWcxhq",{"id":198,"name":199,"description":202,"sessions":205,"colors":428},532,{"en":200,"zh-hant":201},"Open LLM Tech: Core Technologies","Open LLM Tech: 開源模型技術",{"en":203,"zh-hant":204},"This track focuses on the underlying core technologies and development workflows of Large Language Models (LLMs), rather than the generated output. We welcome technical submissions regarding model training, architectural optimization, and deployment practices. Topics include, but are not limited to:\r\n\r\n1. Model Architecture & Training Practices\r\nExplore model structural design, pre-training, and fine-tuning techniques such as LoRA or QLoRA, and methods to improve training efficiency.\r\n2. Quantization & Deployment Optimization\r\nShare insights on model quantization, optimization of inference engines (such as llama.cpp, vLLM), and achieving high-performance deployment across various hardware environments.\r\n3. Dataset Processing & Open Standards\r\nDiscuss the cleaning and labeling processes for high-quality training data, and practices to ensure datasets and model weights comply with open-licensing standards.","本軌聚焦於大型語言模型（LLM）的底層核心技術與開發流程，而非最終生成的內容。我們歡迎關於模型訓練、架構優化與部署實務的技術分享。主題包括但不限於：\r\n\r\n1. 模型架構與訓練實務\r\n探討模型結構設計、預訓練（Pre-training）、微調（Fine-tuning）技術如 LoRA 或 QLoRA，以及如何提升訓練效率。\r\n2. 模型量化與布署優化\r\n分享模型量化技術、推論引擎（如 llama.cpp, vLLM）的優化，以及如何在不同硬體環境中達成高效能部署。\r\n3. 資料集處理與開放標準\r\n討論高品質訓練資料的清洗、標記流程，以及確保資料集與模型權重符合開放授權規範的實務。",{"2026-08-08":206},[207,235,258,287,310,332,354,375,405],{"id":208,"room":209,"start":212,"end":213,"language":214,"track":215,"speakers":217,"zh":225,"en":229,"tags":232,"uri":234},"7Q8NTC",{"en":210,"zh-hant":211},"TR212","","2026-08-08T12:30:00+08:00","2026-08-08T13:00:00+08:00","Mandarin",{"id":198,"name":216},{"en":200,"zh-hant":201},[218],{"id":219,"avatar":220,"zh":221,"en":224},"UDMCG7","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002Fprofile6_PaR8OZx.png",{"name":222,"bio":223},"Martin Chang","Systems software engineer working on HPC, GPGPU, and AI.",{"name":222,"bio":223},{"title":226,"describe":227,"type":228},"可解釋性與應用：人類的最終防線","LLM 與相關技術為科技與人類帶來了巨大的轉變。但我們依然不知道 LLM 內部的原理與如何確保 LLM 不會有惡意行為。可解釋性技術提供了少數的切入點。與其存外部在訓練過程中控制語言模型的行為，不如直接打開模型，去直接探索甚至控制LLM的行為。\r\n\r\n但這麼重要的技術卻鮮少被討論跟應用。這裡我們打開他的面紗，為未來控制更強大的 AI 爭取希望\r\n\r\nCode: https:\u002F\u002Fgithub.com\u002Fmarty1885\u002Fllama.cpp\u002Ftree\u002Frwkv-edit","Talk",{"title":230,"describe":231,"type":228},"Mechanistic interpretability and applications: The final fronteir of hacking","LLM changed the course of tech and humanity for better or for worse. But safety has and still is a problem. No one can guarantee if LLMs are evil or misaligned. Instead of trying to make LLMs safe during training by different means. Mechanistic interpretability provides a different way - try to detect and change LLM behavior by opening them up and see what is going on. \r\n\r\nMechanistic interpretability is an important tech and not an easy one. Yet they are rarely discussed or developed publicly. Let's change that. If not so future us down the road can control much more powerful AI then right now.\r\n\r\nCode: https:\u002F\u002Fgithub.com\u002Fmarty1885\u002Fllama.cpp\u002Ftree\u002Frwkv-edit",[233,214],"Intermediate","https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002F7Q8NTC",{"id":236,"room":237,"start":238,"end":212,"language":214,"track":239,"speakers":241,"zh":249,"en":252,"tags":255,"uri":257},"GC99EF",{"en":210,"zh-hant":211},"2026-08-08T12:00:00+08:00",{"id":198,"name":240},{"en":200,"zh-hant":201},[242],{"id":243,"avatar":244,"zh":245,"en":248},"EDGUYA","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FEDGUYA_wbAgePW.webp",{"name":246,"bio":247},"Tsai,Ming-Han","是個對複雜的事物都很喜歡且很有自己想法的一個人。\r\n\r\nps 我在找工作 有需要都歡迎聯繫!\r\nas95630as@gmail.com",{"name":246,"bio":247},{"title":250,"describe":251,"type":228},"突破 FP64 限制：AdaptiveGEMM 透過 INT8 Tensor Core 在消費級 GPU 逼近 A100 效能","高精度矩陣乘法 (GEMM) 在科學與 AI 運算中是核心算子。然而，消費級 GPU (如 RTX 40 系列) 的 FP64 雙精度算力受硬體限制，成為許多開發者與量化交易的效能瓶頸。\r\n\r\n本議程將分享開源專案 AdaptiveGEMM：探討如何利用實作 Ozaki Scheme 將高精度運算降維拆解。我們將深入底層，利用 CUDA PTX 與 mma.sync 指令極限壓榨 INT8 Tensor Core，解決 Shared Memory Bank Conflict，成功在 RTX 4060 上大幅提升效能，達到逼近 A100 等級的 FLOPS 表現。\r\n\r\n【難易度：進階】適合尋求突破硬體極限的 HPC 與 AI 底層架構工程師。\r\n【先備知識】建議具備基礎 C++ 能力，了解 GEMM 運作原理，並對 GPU 記憶體架構與 Tensor Core 有初步認識。",{"title":253,"describe":254,"type":228},"Breaking FP64 Limits: AdaptiveGEMM Achieves Near-A100 Performance on RTX 4060 using INT8 Tensor Cores","High-precision GEMM is vital for scientific computing and AI. However, consumer GPUs (like the RTX 40 series) are hindered by hardware limitations in FP64 performance.\r\n\r\nThis session introduces \"AdaptiveGEMM,\" an open-source project employing the Ozaki Scheme to algorithmically decompose high-precision operations. We will dive into CUDA architecture, utilizing PTX and mma.sync instructions to maximize INT8 Tensor Core utilization while resolving Shared Memory Bank Conflicts. Discover how this project bypasses hardware restrictions to achieve near A100-level FLOPS on an RTX 4060.\r\n\r\nDesigned for advanced HPC engineers, this talk requires basic C++ proficiency and an understanding of GPU memory hierarchy.",[256,214],"Advanced","https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FGC99EF",{"id":259,"room":260,"start":261,"end":262,"language":263,"track":264,"speakers":266,"zh":281,"en":284,"tags":285,"uri":286},"JRCCHW",{"en":210,"zh-hant":211},"2026-08-08T11:00:00+08:00","2026-08-08T11:30:00+08:00","English",{"id":198,"name":265},{"en":200,"zh-hant":201},[267,274],{"id":268,"avatar":269,"zh":270,"en":273},"B7RWEC","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FB7RWEC_gGdx6YS.webp",{"name":271,"bio":272},"Hrittik Roy","Hrittik is a Platform Advocate at Loft Labs and a CNCF Ambassador, with expertise in cloud native technologies and open source communities. He has contributed extensively to developer advocacy, technical writing, and community engagement. Hrittik has been a featured speaker at events such as Kubernetes Community Days, Open Source Summits, and more, and has served as a Program Committee member for several KubeCons and CloudNativeCons.",{"name":271,"bio":272},{"id":275,"avatar":276,"zh":277,"en":280},"P3JZJE","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FDisplay_Picture_rAx4fAJ.jpeg",{"name":278,"bio":279},"Parth Goswami","Parth Goswami is an open source enthusiast and mostly works with containers and networking. He loves contributing to open source by hosting meetups and interacting with the community.",{"name":278,"bio":279},{"title":282,"describe":283,"type":228},"YAML is the New Dockerfile: Building AI Agent Systems with Docker cagent","The industry is rapidly moving from single LLM applications toward systems composed of multiple agents, tools, memory layers, and workflows. But most discussions around AI agents focus heavily on frameworks and demos, while avoiding the larger systems question:\r\n\r\nWhat does agent infrastructure actually look like?\r\n\r\nThis talk takes a cloud native and systems-oriented view of AI agents. We’ll explore how ideas from distributed systems, containers, orchestration, and platform engineering are beginning to shape the next generation of AI tooling.\r\n\r\nTopics include:\r\n\r\n- Why AI agents resemble distributed systems more than chatbots\r\n- The emerging role of protocols like MCP\r\n- Declarative workflows versus imperative orchestration\r\n- Reproducibility and portability challenges in AI systems\r\n- Why infrastructure concepts like OCI artifacts and YAML workflows are reappearing in AI tooling\r\n\r\nUsing Docker cagent as a concrete example rather than the centerpiece, we’ll examine how modern agent runtimes are evolving toward infrastructure-style abstractions.",{"title":282,"describe":283,"type":228},[233,263],"https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FJRCCHW",{"id":288,"room":289,"start":290,"end":291,"language":214,"track":292,"speakers":294,"zh":302,"en":305,"tags":308,"uri":309},"KCDCT7",{"en":210,"zh-hant":211},"2026-08-08T09:30:00+08:00","2026-08-08T10:00:00+08:00",{"id":198,"name":293},{"en":200,"zh-hant":201},[295],{"id":296,"avatar":297,"zh":298,"en":301},"8KMRCD","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002F8KMRCD_sMRaa5t.webp",{"name":299,"bio":300},"Nero Un 阮智軒","來自澳門的開發者，現職 IBM 顧問，具備豐富的資料科學、資料工程與人工智慧領域實務經驗，曾參與醫療、金融及製造業等跨產業專案，累積多元的產業洞察、解決方案設計與系統開發經驗。 畢業於高雄醫學大學，並取得成功大學醫學資訊研究所碩士學位。曾於生醫新創公司擔任 R & D 及 TPM。 熱衷研究技術探索推動變革的可能，深信科技的力量能為世界帶來影響與改變。 A developer from Macao, currently serving as a Consultant at IBM, with practical expertise in data science, data engineering, and artificial intelligence. Graduated from Kaohsiung Medical University and holds a master’s degree in Medical Informatics from National Cheng Kung University. Previously served as an R & D engineer and TPM at a biomedical startup. Passionate about exploring technology to drive transformative change, and firmly believes in the power of technology to influence and reshape the world.",{"name":299,"bio":300},{"title":303,"describe":304,"type":228},"A Prompt Is Not All You Need：用開源工具打造 LLM 合成資料管線","在開源模型生態蓬勃發展的今天，開發者能輕易取得 Gemma4、Qwen3.5 等頂尖的開源大語言模型 (LLM)。然而，當我們想將這些技術落地到金融、醫療等高合規場景時，往往會撞上一面名為「資料隱私」的高牆。受限於個資 (PII) 與法規，團隊根本拿不到真實業務資料來進行 PoC、效果驗證或 RAG 系統的壓力測試。沒有可用資料進行驗證，再好的開源模型也難以推進到實際的業務中。\r\n\r\n面對「資料可取用性」的難題，合成資料 (Synthetic Data) 成為破局關鍵方法。\r\n\r\n許多團隊最初會嘗試手寫 Prompt 來生成測試資料。但當需求擴增至上千筆，且須同時滿足多樣性、邊界條件與邏輯一致性時，這種做法很快就會面臨品質失控、難以驗證與無法擴展的瓶頸。\r\n\r\n本議程將以 Nvidia Open Source 工具 Data Designer 為例、提供可重現的範例 pipeline，包含 synthetic PII \u002F domain QA seed schema、DAG config、validator 設計、LLM-as-a-Judge rubric，以及生成資料的品質檢查報告。所有範例資料與程式碼將以開源授權釋出，讓聽眾會後可以直接改造成自己的 RAG 測試資料或 PoC dataset。\r\n\r\n透過這個實戰案例，希望幫助開源開發者與企業 IT 團隊在缺乏真實資料的困境中，利用開源工具鏈「無中生有」打造高品質的測試資料集，讓 AI 落地的最後一哩路走得更穩健。",{"title":306,"describe":307,"type":228},"A Prompt Is Not All You Need: Building a LLM Synthetic Data Pipeline with Open-Source Tools","Today, as the open-source model ecosystem flourishes, developers can easily access top-tier open-source large language models (LLMs) like Gemma4 and Qwen3.5. However, when attempting to implement these technologies in highly regulated fields such as finance and healthcare, they often encounter a significant barrier called \"data privacy.\" Due to restrictions related to personally identifiable information (PII) and regulations, teams cannot access real business data for PoC, performance validation, or RAG system stress testing. Without usable data for validation, even the best open-source models struggle to be integrated into actual business scenarios.\r\n\r\nTo address the challenge of \"data accessibility,\" synthetic data becomes a key solution.\r\n\r\nMany teams initially try manual prompt engineering to generate test data. But when the volume increases to thousands of records, and requirements include diversity, boundary conditions, and logical consistency, this approach quickly faces quality control issues, validation difficulties, and scalability bottlenecks.\r\n\r\nThis agenda will take Nvidia's open-source tool Data Designer as an example, providing a reproducible pipeline that includes synthetic PII\u002Fdomain QA seed schema, DAG configuration, validator design, LLM-as-a-Judge rubric, and data quality inspection reports. All sample data and code will be released under open-source licenses, allowing participants to adapt them directly into their own RAG test data or PoC datasets after the session.\r\n\r\nThrough this practical case study, we hope to assist open-source developers and enterprise IT teams in creating high-quality test datasets \"out of thin air\" using open-source toolchains, helping the AI implementation journey be more robust at the final leg.",[233,214],"https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FKCDCT7",{"id":311,"room":312,"start":291,"end":313,"language":214,"track":314,"speakers":316,"zh":324,"en":327,"tags":330,"uri":331},"MMRBXE",{"en":210,"zh-hant":211},"2026-08-08T10:30:00+08:00",{"id":198,"name":315},{"en":200,"zh-hant":201},[317],{"id":318,"avatar":319,"zh":320,"en":323},"VWXHUR","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FVWXHUR_aiYkdjx.webp",{"name":321,"bio":322},"galaxy4552","Hi, I'm galaxy4552, an independent developer exploring alternative approaches to language models and semantic retrieval.\r\nThis talk simply shares some discoveries from my experiments with the open-source project PipeOwl.\r\n\r\n**_The Universe Speaks in Numbers._**",{"name":321,"bio":322},{"title":325,"describe":326,"type":228},"另一種 Encoding 方法：將語意檢索從O(n²)變成O(n)","目前多數語意搜尋與 embedding 系統，通常依賴 Transformer 或向量相似度計算，其複雜度常接近 O(n²)。在大型語料或即時搜尋場景中，這會帶來顯著的計算成本。\r\n\r\n本分享將介紹一種不同的 encoding 思路，透過結構化語意表示與詞彙場（semantic field）的概念，將查詢與詞彙評分的複雜度降低至 O(n)。\r\n\r\n演講內容包含：\r\n\r\n- 傳統 embedding \u002F Transformer 檢索的計算瓶頸\r\n- 一種替代的語意 encoding 架構\r\n- PipeOwl 專案的設計與實作\r\n- 多語言模型（中文 \u002F 日文）的實驗結果\r\n- 與 BM25、Embedding、FAISS 的速度比較\r\n\r\n本方法與模型已開源，並提供實際 benchmark 與實作程式碼，期待與開源社群討論另一種語意檢索的可能方向。\r\n\r\n相關專案\r\n- https:\u002F\u002Fhuggingface.co\u002FWangKaiLin\u002FPipeOwl-1.6-tw\r\n- https:\u002F\u002Fhuggingface.co\u002FWangKaiLin\u002FPipeOwl-1.5-jp\r\n- https:\u002F\u002Fhuggingface.co\u002FWangKaiLin\u002FPipeOwl-1.4-multilingual",{"title":328,"describe":329,"type":228},"An Alternative Encoding Method: Reducing Semantic Retrieval from O(n²) to O(n)","Modern semantic retrieval systems are typically built on Transformer embeddings and vector similarity search, which often result in near O(n²) computational behavior.\r\n\r\nThis talk introduces PipeOwl, an alternative encoding method that treats semantic representation as a structured scoring field over vocabulary, enabling O(n) query evaluation.\r\n\r\nI will discuss the design motivations, implementation details, and benchmark results comparing PipeOwl with BM25, traditional embeddings, and FAISS-based retrieval systems.\r\n\r\nThe project is fully open source and includes multilingual models for Chinese and Japanese.",[233,214],"https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FMMRBXE",{"id":333,"room":334,"start":262,"end":238,"language":214,"track":335,"speakers":337,"zh":345,"en":348,"tags":351,"uri":353},"S8MFLU",{"en":210,"zh-hant":211},{"id":198,"name":336},{"en":200,"zh-hant":201},[338],{"id":339,"avatar":340,"zh":341,"en":344},"TLWDMZ","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FTLWDMZ_BN7DNfj.webp",{"name":342,"bio":343},"Max","目前就讀於成功大學資訊工程系",{"name":342,"bio":343},{"title":346,"describe":347,"type":228},"在 FPGA 上從 RTL 到 Token 走過 LLM 推理全流程","我透過 DE10-Nano FPGA 開發板，沒有 GPU、不呼叫任何框架，從模型量化、推論引擎到 RTL 硬體設計，在 AI 推理的每一層都讓自己做中學，將 BitNet 2B transformer 模型塞在 1GB 記憶體的 DE10-nano FPGA 上完成 token 輸出。\r\n\r\n此演講適合那些想要摸清 LLM 從底層硬體到最終 token 輸出的人。\r\n\r\n這場分享會聊：三元權重 LLM 為何特別適合 FPGA、T-MAC 查表的實作與資源平衡、HPS\u002FFPGA 混合推理引擎的設計取捨、以及在資源極度受限硬體上做 LLM 部署的踩坑經驗。",{"title":349,"describe":350,"type":228},"From RTL to Token: Walking Through the Complete LLM Inference Pipeline on an FPGA","Using a DE10-Nano FPGA development board—with no GPU and no frameworks—I built every layer of the AI inference stack myself as a learning-by-doing journey: from model quantization and the inference engine down to RTL hardware design, ultimately getting a BitNet 2B transformer model to generate tokens on the DE10-Nano.\r\nThis talk is for anyone who wants to understand how an LLM works all the way from the underlying hardware to the final token output.\r\nTopics covered: why ternary-weight LLMs are a natural fit for FPGAs, implementing T-MAC lookup tables and balancing hardware resources, design trade-offs in the HPS\u002FFPGA hybrid inference engine, and lessons learned from deploying an LLM on extremely resource-constrained hardware.",[352,214],"Elementary","https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FS8MFLU",{"id":355,"room":356,"start":313,"end":261,"language":263,"track":357,"speakers":359,"zh":367,"en":370,"tags":373,"uri":374},"SVZZQG",{"en":210,"zh-hant":211},{"id":198,"name":358},{"en":200,"zh-hant":201},[360],{"id":361,"avatar":362,"zh":363,"en":366},"KWSYWJ","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FKWSYWJ_sh4wHup.jpg",{"name":364,"bio":365},"Tommy Han","Tommy is a software engineer in Hong Kong who loves to work on open source projects in different areas, including toolchains, compilers, apps, and IDE development.",{"name":364,"bio":365},{"title":368,"describe":369,"type":228},"從 PyTorch 到硬體：wait what? 甚麼是 LLM 編譯器...?","大型語言模型（LLM）的開發雖然可能始於 PyTorch 等框架，但要讓它們在 CPU、GPU 及專用加速器上高效運行，越來越依賴於編譯器基礎架構（Compiler Infrastructure）。\r\n\r\n隨著機器學習系統變得日益複雜，硬體也更趨向於專用化，編譯器層已成為連接模型程式碼與機器執行之間不可或缺的關鍵橋樑。\r\n\r\n本場演講將從實用角度介紹 AI 編譯器，涵蓋它們為何至關重要、能解決哪些類型的問題，以及 **LLM MLIR** 和 **OpenXLA\u002FXLA** 等開源專案在整個技術棧（Stack）中扮演的角色。此外，我們也將簡要介紹 **tinygrad**，它正嘗試採用不同於 LLVM 的另一種路徑，來編譯並在硬體上運行模型。\r\n\r\n演講中也將解釋**中間表示（Intermediate Representation, IR）**、降低（Lowering）**以及**後端目標定位（Backend Targeting）等概念，是如何將 PyTorch 等框架與實際硬體連接起來的。\r\n\r\n最後，我們將透過一個基於 PyTorch 與多種編譯器（例如 LLVM 和 tinygrad）的簡單 Demo，讓整個管線（Pipeline）的運作流程變得更具體、更容易理解。",{"title":371,"describe":372,"type":228},"From PyTorch to Hardware: An Introduction to LLM Compilers","Large Language Models may start in frameworks like PyTorch, but getting them to run efficiently on CPUs, GPUs, and specialized accelerators increasingly depends on compiler infrastructure.\r\n\r\nAs machine learning systems have grown more complex and hardware has become more specialized, the compiler layer has become a key bridge between model code and machine execution.\r\n\r\nThis talk introduces AI compilers from a practical perspective, covering why they matter, what kinds of problems they solve, and how open-source projects such as LLM MLIR and OpenXLA\u002FXLA fit into the broader stack. \r\nIt will also introduce a bit about tinygrad, which is trying to take another approach rather than LLVM, to compiler models running on the hardware..\r\n\r\nIt also explains how ideas such as intermediate representations, lowering, and backend targeting connect frameworks like PyTorch to real hardware.\r\n\r\nA small demo based on PyTorch and various compilers (e.g. LLVM and tinygrad) helps make the overall pipeline more concrete and easier to understand.",[256,263],"https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FSVZZQG",{"id":376,"room":377,"start":378,"end":379,"language":263,"track":380,"speakers":382,"zh":397,"en":400,"tags":403,"uri":404},"UEUXUA",{"en":210,"zh-hant":211},"2026-08-08T14:00:00+08:00","2026-08-08T14:30:00+08:00",{"id":198,"name":381},{"en":200,"zh-hant":201},[383,390],{"id":384,"avatar":385,"zh":386,"en":389},"TB8FK8","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002FTB8FK8_gmlbtw3.webp",{"name":387,"bio":388},"Mario Behling","Mario Behling is the co-founder of FOSSASIA and a member of the Visdom team. He works on open source projects, developer communities, and open technology events across Asia and beyond. At FOSSASIA, he supports projects such as Visdom, eventyay, PSLab, and other open source tools, with a focus on community collaboration, sustainable development, and practical open source infrastructure.",{"name":387,"bio":388},{"id":391,"avatar":392,"zh":393,"en":396},"XWW9AQ","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002F6468f3df-ff33-4e45-b2e6-f17576f24ddc_fChltVe.jpg",{"name":394,"bio":395},"François Cartegnie","I'm here to set up the remote session for the talk.\r\nI belong to another unrelated project.",{"name":394,"bio":395},{"title":398,"describe":399,"type":228},"From Experiments to Dashboards: Modernizing Visdom for Open AI Workflows \u002F 從實驗到儀表板：為開放 AI 工作流程現代化 Visdom","Visdom is an open source tool for live visualizations and lightweight dashboards in Python, PyTorch, NumPy, and AI workflows. In this session, François Cartegnie will facilitate the presentation and conversation with Mario Behling and the Visdom team which will share the story of Visdom’s modernization at FOSSASIA, including the newest release, local deployment, and an emerging online version that lets researchers and developers quickly use Visdom for their projects.\r\n\r\nVisdom 是一套開源工具，可用於 Python、PyTorch、NumPy 和 AI 工作流程中的即時視覺化與輕量級儀表板。在本場次中，Visdom 團隊的 Mario Behling 將分享 Visdom 在 FOSSASIA 推動現代化的故事，包括最新版本、本機部署，以及正在發展中的線上版本，讓研究人員和開發者能更快速地將 Visdom 用於自己的專案。",{"title":401,"describe":402,"type":228},"From Experiments to Dashboards: Modernizing Visdom for Open AI Workflows","AI and scientific computing workflows produce metrics, plots, images, logs, model outputs, and many other experimental results. Developers and researchers need practical ways to inspect this data, compare experiments, and share insights without depending only on closed or heavyweight platforms.\r\n\r\nVisdom is an open source tool for creating, organizing, and sharing live visualizations of rich data. Originally developed at Facebook AI Research and later transitioned to FOSSASIA, Visdom has a long history and an existing user base in Python, PyTorch, NumPy, and scientific computing communities.\r\n\r\nIn this session, François Cartegnie will facilitate the presentation and conversation with Mario Behling and the Visdom team that shares the story of Visdom’s next phase: how the project is being modernized for current AI workflows, what kinds of technical and community challenges come with maintaining an established open source tool, and why open visualization tools remain relevant in today’s AI ecosystem.\r\n\r\nThe talk will cover two important directions for Visdom: local-first deployment for developers who want to run the tool in their own environment, and a new online version that enables researchers and developers to quickly upload or connect their projects and start using Visdom without setting up their own instance first.\r\n\r\nAt COSCUP, we will also present the newest release of Visdom and discuss current development directions, including usability, frontend and backend modernization, Python client improvements, documentation, testing, and new visualization features. The session invites COSCUP participants to explore how open source communities can keep important AI tools useful, sustainable, and open.\r\n\r\nThe session will be held in a mix of English and Chinese, with team members helping with translation to make it accessible to both international and local audiences.",[233,263],"https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FUEUXUA",{"id":406,"room":407,"start":408,"end":378,"language":263,"track":409,"speakers":411,"zh":419,"en":423,"tags":426,"uri":427},"ZVVRHF",{"en":210,"zh-hant":211},"2026-08-08T13:30:00+08:00",{"id":198,"name":410},{"en":200,"zh-hant":201},[412],{"id":413,"avatar":414,"zh":415,"en":418},"RYNSZY","https:\u002F\u002Fpretalx.coscup.org\u002Fmedia\u002Favatars\u002F%E4%B8%8B%E8%BC%89_zKOUUfZ.png",{"name":416,"bio":417},"John Lu","John is a Senior Machine Learning Engineer from a Top-tier company, currently focused on developing NLP applications.\r\n\r\nHe is deeply motivated by challenges and tends to be excited by breaking conventional ways of thinking and doing. With prior experiences in Software Engineering, he works on combining the latest AI technology and engineering to transform challenges into practical solutions.",{"name":416,"bio":417},{"title":420,"describe":421,"type":422},"使用 PyTorch 從零量化 YOLOX","模型量化 (Model Quantization) 能在推論階段中有效降低計算量與記憶體使用量。 透過將模型權重與輸入資料轉換為低精度資料型態，不僅能加快運算速度，還能讓模型部署在嵌入式裝置。 本次演講將示範如何以 Python 實作模型量化。我們將從零開始，使用 PyTorch 對 YOLOX 物件偵測模型進行量化，帶領大家從原理到實作，將量化技術應用於開源模型，並轉化為可落地的解決方案。","workshop\u002Fpanel",{"title":424,"describe":425,"type":422},"Let's Quantize YOLOX from scratch with PyTorch","Model Quantization can reduce the computational and memory footprint during inference.  The weights and activations are represented with low-precision data types and can operations can be performed faster. Model Quantization allows us to run models on embedded devices. The goal of this talk is to demonstrate how Model Quantization could be implemented in a Pythonic way. To do so, we're going to quantize the YOLOX object detection model completely from scratch all using PyTorch.",[256,263],"https:\u002F\u002Fcoscup.org\u002F2026\u002Fsession\u002FZVVRHF",{"2026-08-08":429},"#264653",[431,436,440,444,447,450,454,458,462,467],{"id":432,"imageVertical":433,"imageHorizontal":434,"link":435,"weight":44},"QNAP-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1-HQ9oQMOngtqM3yL-1DBHRhzWbuyxgfE&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1aVhw014dwVLuxbNadE314lnl992FcjS_&&sz=w4000","https:\u002F\u002Fwww.qnap.com\u002Fgo\u002Fsolution\u002Fvirtualization",{"id":437,"imageVertical":438,"imageHorizontal":439,"link":74,"weight":163},"Cathay-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=10ndsUHILq9m_boaoilHTERyTXCsCBlSX&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=11_VlhFrCYcNOHML4z0jm6AmyL7MxNTfl&&sz=w4000",{"id":441,"imageVertical":442,"imageHorizontal":443,"link":111,"weight":163},"esunbank-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1aNURMNN83oIZHAtBzm2rUZODKVBO29dt&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=17OA0eu-N510ePhhDuiSuBV_4mdlU5gfi&&sz=w4000",{"id":211,"imageVertical":445,"imageHorizontal":446,"link":152,"weight":163},"https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1_NsnJt2O238fO_dLl19RR4SNbyRErIJY&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1Q989E0GWazdVSPcGvZLMYzsNSh4cV2-g&&sz=w4000",{"id":211,"imageVertical":448,"imageHorizontal":449,"link":162,"weight":163},"https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1d4-1ZBMs6GqYcYytlYrSVZhs8Jgye5l3&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1Ea4CNZ-14siicP3xYm78l3AF41iAb9Om&&sz=w4000",{"id":211,"imageVertical":451,"imageHorizontal":452,"link":453,"weight":163},"https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1A9Q35aOO4LDtI1-MVbIlLoGBmWn2VwZg&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=13Oz8Eq6MXXspGEqwi5QGbwYyDNeU1idK&&sz=w4000","https:\u002F\u002Fnitra.com\u002F",{"id":455,"imageVertical":456,"imageHorizontal":457,"link":55,"weight":44},"kktix-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=11md6cNzKBG-OF_e_9G5ZCLujoyck96as&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1ec5tuBMqbKDfPakOPWZF93Q55JI7DOOl&&sz=w4000",{"id":459,"imageVertical":460,"imageHorizontal":461,"link":122,"weight":22},"gamesofa-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1HkRfiNOhKrH2zTaOHjTkDnAmoDPqV5_c&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1IVAroDzzryU5AIY4qMMVztg4hu-FXaIM&&sz=w4000",{"id":463,"imageVertical":464,"imageHorizontal":465,"link":466,"weight":44},"onlyoffice-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1H7dguSsKUH9aS9vxK-RhI519LEQjaak-&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1IchePy7nwLT4TXnLYLf8o6HS4TKOOjFd&&sz=w4000","https:\u002F\u002Fwww.onlyoffice.com\u002F",{"id":468,"imageVertical":469,"imageHorizontal":470,"link":180,"weight":44},"SUN-SQUARE-news","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=19TU_W9JBLzrRYVSQPG4cbbckhWLy0JaS&&sz=w4000","https:\u002F\u002Fdrive.google.com\u002Fthumbnail?id=1-2rlsJngO4Qiu0SElyXhMYmsjmGokHu_&&sz=w4000",{"data":472,"body":473},{},{"type":474,"children":475},"root",[476,484],{"type":477,"tag":478,"props":479,"children":480},"element","p",{},[481],{"type":482,"value":483},"text","This track focuses on the underlying core technologies and development workflows of Large Language Models (LLMs), rather than the generated output. We welcome technical submissions regarding model training, architectural optimization, and deployment practices. Topics include, but are not limited to:",{"type":477,"tag":485,"props":486,"children":487},"ol",{},[488,494,499],{"type":477,"tag":489,"props":490,"children":491},"li",{},[492],{"type":482,"value":493},"Model Architecture & Training Practices\nExplore model structural design, pre-training, and fine-tuning techniques such as LoRA or QLoRA, and methods to improve training efficiency.",{"type":477,"tag":489,"props":495,"children":496},{},[497],{"type":482,"value":498},"Quantization & Deployment Optimization\nShare insights on model quantization, optimization of inference engines (such as llama.cpp, vLLM), and achieving high-performance deployment across various hardware environments.",{"type":477,"tag":489,"props":500,"children":501},{},[502],{"type":482,"value":503},"Dataset Processing & Open Standards\nDiscuss the cleaning and labeling processes for high-quality training data, and practices to ensure datasets and model weights comply with open-licensing standards.",1785169686223]