* Co-founded and led a pioneering research group focused on Japanese Language LLM evaluationSpearheaded the development of Rakuda, a comprehensive evaluation framework for Japanese LLMs:* Designed and implemented a multi-faceted assessment methodology covering various aspects of Japanese language understanding and generation* Created benchmarks for evaluating both commercial and open-source LLMs in Japanese language tasks* Established and maintained partnerships with major corporations like Toyota and Amazon Japan, to implement and refine Rakuda, continuously improving its relevance and effectiveness through industry feedback* Led a team of researchers and engineers in ongoing refinement and expansion of the Rakuda framework

Led technical development for an AI fintech startup focused on the Japanese financial market. Key responsibilities and achievements include:* Architected and implemented an advanced custom Retrieval-Augmented Generation (RAG) system for financial analysis and decision-making, built from scratch without relying on frameworks like LangChain.* Designed a sophisticated multi-stage retrieval pipeline integrating dense and sparse retrieval methods, custom embedding models, and a novel re-ranking algorithm.* Developed a dynamic context selection mechanism and integrated real-time market data feeds to ensure up-to-date and optimized information retrieval for financial queries.* Created innovative solutions to enhance Large Language Model (LLM) performance for Japanese language processing, including fine-tuning on financial datasets and custom tokenization techniques.* Increased LLM accuracy for Japanese financial text analysis by 22% and improved relevance of retrieved documents by more than 30% compared to baseline RAG systems.* Collaborated with financial professionals to integrate LLM-driven solutions, resulting in over 20% reduction in work hours for end-users.* Played a key role in product strategy, aligning technical capabilities with market needs and business objectives.

* Full-stack developer lead on a distributed scalable React with Hooks web application used for AI facial recognition solutions. This project brings end-to-end solution to small/medium size organizations without the need of in-house developer, and generate new revenue source.* Perform as full-stack developer for AI age verification/detection web app, which adopted by sales team to demonstrate AI technology and lead to ongoing talk with Fortune 100 companies.* Coached and mentored junior software developers and held training sessions and pair programming sessions to facilitate knowledge sharing, professional growth, quality of code, and ultimately maintain the speed of development to meet the company and clients' demand. * Adopt Jest/Enzyme test to ensure JS/React functionality, and Cypress for end-to-end testing to provide complete test coverage in TDD environment. In addition with Circle CI to optimize and automate code base maintenance.

• Developed (using NodeJS runtime) an Electron app for RabbitMQ message broker to help communicate insights derived from analyse data to client that easily identifies network infrastructure, bottlenecks, which allows for faster decision making.• Leveraged React and D3 to build a dynamic and interactive visualization based on its data-driven DOM manipulation to translate data into easily digestible, simple and powerful visual story for developers to view RabbitMQ instance in bird eye view.• Built React reusable modular components to render RabbitMQ message and publish rate while maintaining a consistent, easy to maintain/grow codebase to reduce development cycle.• Designed a modern user interface with Material UI, leveraging production grade components for faster development.• Deployed multiple RabbitMQ instances to serve as a foolproof sandbox for application development to test and facilitate high volume data transfer between clients and servers within application. • Adopted Travis CI to optimize deployed codebase updates and automated Jest, Enzyme test to ensure JS/React functionality follows TDD best-practices.• Packaged and delivered cross-platform Electron app to enhance security, reusability, update, code and app management.(Open Source project at accelerated CS Lab)

- Architected and engineered with JavaScript and jQuery news website to aggregate and render article and post data.- Developed SQL workflow to streamline process, create, update and delete content via frontend user interface.- Integrated analytics library to identify traffic trends, pathing issues, opportunities for improving application UX.