Selected Projects

Tech Content — Graduation Thesis

Next.jsTypeScriptPythonFastAPIOpenAI APIClaude APICloudflareSupabase

My graduation research explores how AI-coding articles on Qiita relate to community response.

Based on the findings, I built an AI writing tool that helps authors plan, generate, and edit technical articles while keeping them in control of the final content.

CraftMend

Frontend Demo Only
Next.jsTypeScriptPythonFastAPISupabaseGoogle Cloud

Built for Agent Forge AI Hackathon: Tokyo Creativity, a one-day AI agent hackathon held at Mercari's Tokyo office on July 25, 2026.

CraftMend helps owners of damaged Japanese craft items find a path to repair. From a single photo, it produces a confidence-aware damage assessment, searches for real repair studios, ranks the results against a human-verified dataset, and drafts a Japanese inquiry after the user chooses which studio to contact.

The pipeline is designed around trustworthy automation: model outputs are schema-validated, data provenance is preserved, unknown information remains unknown, external failures degrade visibly, and no inquiry is sent without explicit human approval.

This project is currently available as a frontend-only demo. Backend services are paused, and no functionality is currently available.

Living Protocol Copilot

Frontend Demo Only
Next.jsTypeScriptPythonFastAPIClaude APICloud Run

Built for the Builder Track of Built with Claude: Life Sciences, a global virtual hackathon hosted by Anthropic and Cerebral Valley in partnership with Gladstone Institutes in July 2026.

Living Protocol Copilot turns existing wet-lab SOPs into structured, risk-annotated living protocols and provides source-grounded troubleshooting linked to both the user's protocol steps and curated published references. It highlights missing information, communicates uncertainty instead of guessing, and remains strictly a decision-support tool for qualified review.

This project is currently available as a frontend-only demo. Backend services are paused, and no functionality is currently available.

MediRoute

ReactGemini APIGoogle CloudCloud Run

An AI health agent for foreign travelers in Japan, built for the Gemini AI Hackathon at Google Japan on June 27, 2026, where our team was selected as a Top 10 finalist.

MediRoute supports the medical journey from symptom guidance to clinic search, live Japanese communication support, medicine explanation, and travel-insurance claim preparation while keeping guidance clearly separate from medical diagnosis.

MediRoute is designed as a mobile app. For the best experience, open the live app on a smartphone.

AI Daily & News

Next.jsGemini APIAWS

A daily AI briefing product that curates updates from major AI labs, research archives, open-source communities, and GitHub trending repositories.

The service uses Gemini to generate a concise editorial summary and delivers it by email at 08:00 JST, with subscription, terms, and privacy flows in place.

Japanese Learning Platform

Next.jsGitHub Pages

A personal learning site for organizing Japanese study materials from JLPT N5 toward N1, starting with pronunciation, grammar, and vocabulary lessons.

The site reflects my own study background in Japan and packages lesson content into a structured course navigation experience.

Shift Agent

PythonTypeScriptReactLangChainFastAPIGoogle Cloud

An AI shift management agent developed for the Google Cloud AI Hackathon 2.0.

The system uses Gemini and LangChain to generate, evaluate, and revise shift schedules, with a backend designed around maintainable domain logic.

Text to 3DCG Simulation

Unreal EngineC++PythonBlueprintOpenAI API

This project is a text to 3DCG simulation project. I used the Unreal Engine 5.3 to create a 3DCG simulation.

In this project, I proposed a new method to generate 3DCG from text. This project is about the tatics of basketball. Recently, people are using the whiteboard to draw basketball tatics. However, this method is not efficient and it is not easy to understand. New basketball players especially struggle to visualize plays and movements from static whiteboard drawings. To solve this problem, I developed a system that converts text descriptions of basketball into 3D visualizations. Users can input tactical instructions using natural language, and the system automatically generates an animated 3D scene showing player movements and play development. And I also write a paper about this project.

Zenn Hackathon Note Summarizer

React.jsFastAPIPostgreSQLGemini API

A team project focused on summarizing notes with Gemini API.

I led the project and developed a full-stack application using React.js, FastAPI, and PostgreSQL.