Kanta Y.

Experience

• 

  Fukuoka Directive Council (福岡地域戦略推進協議会, Fukuoka D.C., FDC)

  Aug 2019 - Aug 2019

  Summer Intern at Fukuoka’s Municipal Think-Tank

  Throughout an internship opportunity by "Multicultural Venture Life Challenge 2019", we—students from all over the world and Japan—worked on problem-based learning in the real world. Our group surveyed Fukuoka's inbound tourism and proposed a revitalization plan for a rural place with the potential of rich hot springs over two weeks in Fukuoka, Japan. Takeaways from the program: design thinking, verbal and written business communications in English, data analysis—including primary data collection, which was one of the most challenging work for us. Finally, we gave a presentation at Fukuoka Growth Next. I met amazing people from all over the world and learned how business people strategically carry out surveys and propose ideas based on qualitative/quantitative manners.*This internship was led by Kyushu Univ, OPU, NAIST, Chalmers University of Technology, and others from public/academic sectors. *FDC is Fukuoka's municipal think-tank that collaborates with public, private, educational, and civic sectors. 部分表示

• 

  Self Employed

  Sept 2019 - Dec 2021

  Dating App Development & Business Plan Creation (Web, Mobile)

  This student project involved conception, ideation, prototyping, and putting the prototype into the market. Due to their UI design, today's dating apps in the mainstream focus on our appearances, and they fail to consider that our personality counts when considering long-term relationships. With this in mind, our university students have created new dating apps with unique user experiences emphasizing our inner character rather than superficial lookism. The app initially hid a user's facial photo with a blur when a pair of users matched. Throughout the message, a couple of users can know each other, and the quantity of interaction gradually increases its "intimacy scores," revealing a user's facial image.Although today the app and project itself are no longer active, my main contributions to our team were: (1) the development of API endpoints and management dashboard interacting Firebase/GCP, (2) giving a talk or presenting our work in pitch events, (3) writing a technical case study for the tech community regarding BaaS (Firebase), at the time case studies were insufficient among Japanese tech communities. I wrote tech articles about how our team developed our software utilizing each building block of Firebase in detail—with the assumption that detailed case studies of app development flow using Firebase weren't actively shared by Japanese tech communities; As a result, my article got approx. 20000 views, 103 LGTM, and 76 Stock on a tech blog platform called Qiita, attracting new users of—mostly tech people— after releasing the app. (The article: https://qiita.com/kanta_yamaoka/items/7ecdd3c57206e4aff3f0 )As for (2), I have attended several pitches (My talk is available at https://www.youtube.com/watch?v=ec8mAN0rcgE&t=1432s ) and won a prize (See AWARDS section.) 部分表示

• 

  株式会社ロッケン - Rokken

  Jan 2021 - Aug 2022

  Machine Learning Engineer

  At http://rokken.tech, I supported our team with an on-prem ML infrastructure for model training/inference, and software development to deliver deep learning into production. I joined the legally compliant R&D process for a regulated and safe-critical system which required technical documentation to validate my contributions. Mentors: Thomas Henn & Clément Jacquet. Keywords: IaC, MLOps, Docker, Ansible, Python, OpenCV (C++)Major accomplishments: (i) Improved deep learning GPU cluster environment with IaC, and containerization. My automation—from provisioning new nodes to deploying new updates to workload managers inside docker containers—replaced previous repetitive manual operations to each on-prem node ( approx. 30 machines) done by research engineers with Ansible and bash scripts running via cron. Now they can concentrate on sheer research, not being distracted by repetitive shallow work coming from cluster management.(ii) Engaged in developing specific modules for our project using C++, OpenCV, and Boost. This includes coding with Google C++ Style Guide, and documentation as a process of R&D. I was responsible for making a specific module that interacts with main software that is time-sensitive and mission-critical. The final version has become faster than my first dirty implementation of the module by improving the data structure of graphs. Also, to deal with the memory leak issue of C++, I set up the Gitlab CI pipeline to detect it before the code is merged to the main repository.(iii) Engaged in model performance comparison with GPU in mind using PyTorch and cudaEvent API to suggest promising computer configurations with specific types of models, and inference batch sizes, etc.(iv) Engaged in terabyte-scale dataset management. This includes creating a periodic backup script with rsync, carrying out data migration from old NAS to a new one, re-installing HDD, and merging specific types of datasets that meet specific conditions using Python scripts. 部分表示

• 

  Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI)

  Mar 2022 - Mar 2023

  Research Assistant Intern

  The one-year research intern was funded by JASSO scholarship and research assistant role (Hilfswissenschaftler) at DFKI under Prof. Andreas Dengel, Prof. Shoya Ishimaru. Subgroup: Immersive Quantified Learning Lab in Smart Data & Knowledge ServicesI worked on a learning personalization system, which combines social media pictures and a large-language model (GPT-3) to personalize a person's English learning journey―to find vocab tied to a learner's experience. The initial version was presented in an int’l conference with peer review process: Kanta Yamaoka et al. "Experience is the Best Teacher: Personalized Vocabulary Building Within the Context of Instagram Posts and Sentences from GPT-3" - https://doi.org/10.1145/3544793.3560382 - 9 citations on ResearchGate, Research Interest Score is higher than 87% of items published in 2022 on ResearchGate, 180 views on ACM Digital Library.Our system used their Instagram images to get object labels, generating sentences and displaying them altogether where a learner can find unknown vocab. It consists of Google Cloud ML services such as Vision, NLP (morphological analysis to extract lemma or part-of-speech), and Translation API. It also used OpenAI’s completion API. Final version also integrated Instagram API, Cloud Run/Auth/Firestore/Storage/Function to allow a user to import their images, wait for context creation on background, and proceed to study and test sessions seamlessly. The majority of the containerized backend on Cloud Run was written in Python; The front-end was developed with TypeScript React and hosted into Vercel.Side project: I helped implement an experiment system for a student project in the cognitive science dept in RPTU. It involved real-time data collection from an eye-tracking device - Tobii Eye Tracker 4C, and was developed with React (TS) and Electron, and connected the device with websocket, including instructions for a user, calibration for the tracker, and reaction-time tasks and reading tasks. 部分表示

• 

  Google

  Apr 2023 - now

  Datacenter Technician

  I keep Google's hyperscale clusters up and running in the Hardware Operations team within Technical Infrastructure.I aim to deliver reliable and effective operations for long-term benefits to org and customers using Cloud, constantly seeking for process improvement with 10x! mindset. I always try to extract valuable data from large data sources and make it useful for daily decisions to save my minutes and my colleagues'. I try to share solutions with our global teams in usable manners (e.g., documentation or easy-to-use scripting, tooling, etc.).Major accomplishments:(i) Designed an onsite career program for students: “Google Data Center Hardware Hackathon 2023”. The event was held in Japan and Australia to encourage students to know more about careers at data centers and Google. The program provided practical hands-on problem solving activities for students who are interested in data center infrastructure in Japan and Australia. As part of the international workgroup, I helped design the program for two sites, prepared servers and necessary hardwares for the Japan site, facilitated the event sessions with other Googlers in Japan, and documented the process, sharing insights to the Australian team. I could only be onsite for the Japan event, but both Japan and Australia events went successful without major issues. Key responsibilities:(1) Deployment: deliver high-performance servers and network infrastructure, including highly complex routers, switches, etc.(2) Daily maintenance: troubleshoot servers/network infrastructures from hardware-level to software-level, often escalating blockers to cross-functional teams, such as SWEs, SREs, or PgMs to meet our SLOs.(3) Escalated infrastructure maintenance: troubleshoot mission critical infrastructure, such as facilities or network infrastructure.(4) Process improvement: suggest feedback to internal tools, documentation and pilot experimental projects to improve our organization locally and globally. 部分表示