🇯🇵 README_ja.md

I'm a backend engineer with strengths in data analysis and AI applications.

I have experience developing systems that utilize natural language processing, generative AI, and cloud services,
and have driven projects both independently and within teams.

Outside of work, I build civic tech map services using local government open data
and web applications powered by generative AI, and keep practicing competitive programming on AtCoder.

• Category: Business System
• Responsibilities: Design, Coding, Testing, Data Analysis
• Role: Backend Engineer
• Technologies:

Developed a PoC tool that generates accurate kana readings from medical research texts for use in text-to-speech systems.
To handle domain-specific terminology and complex word variations in medical documents, I analyzed sample data and referenced overseas Japanese linguistic research resources to identify phonetic patterns.
Built the tool in Python, integrating MeCab with custom dictionaries and rule-based logic.
Achieved 99% reading accuracy, successfully completing the PoC.

• Category: Web Service / In-house Product
• Responsibilities: Design, Coding, Testing, Operation & Maintenance, Data Analysis
• Role: Backend Engineer
• Technologies:

Developed a generative AI system that provides precise and fast answers to user queries based on document data.
Implemented RAG (Retrieval-Augmented Generation), which significantly improved response accuracy and contextual relevance.

• Category: Web Service / In-house Product
• Responsibilities: Design, Coding
• Role: Backend Engineer
• Technologies:

Developed a system designed to generate high-quality images based on user-provided text using StableDiffusion.
As a backend engineer, I was responsible for managing the model invocation and generation workflow.
Additionally, implemented an AI-based text validation module to enhance the overall quality of generated images.

An outing navigator for seniors and people with limited mobility in the Tama area of Tokyo. Shows routes where the Tokyo Silver Pass can be used, along with barrier-free facilities.

An unofficial site that explains bills under deliberation in the Aichi Prefectural Assembly and Nagoya City Council in plain language, without jargon.

Determines area risk levels, expected flooding, and evacuation shelters from an address.
Includes a simulator for how many days residents can shelter at home, and a board that matches supply needs with offers.

Just pass a GitHub repository URL to automatically generate multiple-choice quizzes and task-oriented reference docs from the actual code. I also use it to get up to speed with unfamiliar languages and codebases.

Turns the questions and answers in a meeting recording into a mind map. Gemini checks whether each answer actually addresses what was asked, visualizing miscommunication in meetings.

More projects are listed in my Notion portfolio.