📍 Tokyo, Japan · 🤖 NLP Engineer · 🔧 AI Agents & Harness Engineering

I’m an NLP Engineer with a background in natural language processing, knowledge graphs, and LLM applications.

More recently, I’ve been exploring AI Agents and Harness Engineering, building tools, interfaces, and workflows that make agents easier to steer, inspect, and use in real-world tasks.

My current interests include agentic workflows, developer tooling, human-in-the-loop systems, and building practical agent applications.

• 🤖 AI Agents
• 🔧 Agent Harnesses & Developer Tooling
• 🔄 Human-in-the-loop Agent Workflows
• 🧩 Agent UX & Tool Interfaces
• 🧠 LLM & NLP Applications
• 📝 Japanese Language Processing

I'm especially interested in the layer around the model: how agents use tools, receive feedback, manage workflows, and interact with humans.

• 📊 pydata-notebook: Chinese translation notes and examples for Python for Data Analysis, 2nd Edition.

• 🐍 Automate-the-Boring-Stuff-with-Python-Solutions: Solutions and learning notes for Automate the Boring Stuff with Python.

• 📐 statistical-learning-method: Study notes and Python implementations based on Li Hang's Statistical Learning Methods.

• 📚 knowledge-graph-learning: A curated collection of knowledge graph tutorials, projects, papers, tools, and communities.

• 🧠 nlp-beginner-guide-keras: NLP model implementations with Keras for beginners, covering text classification, embeddings, sequence labeling, and other fundamental NLP tasks.

• 📰 news-graph: Key information extraction from text and graph visualization.

• 🕸️ sanguo-kgqa: A knowledge graph based visualization and question answering system for character relationships in Romance of the Three Kingdoms.

• 🎬 word2vec-movies: A Python 3 tutorial for Kaggle's Bag of Words Meets Bags of Popcorn, covering Word2Vec and movie review classification.

• 💬 sentiment-analysis: Sentiment analysis experiments and implementations for text classification.

• 🎯 dsh-annotate: Visual browser feedback for DeepSeek Harness. Select UI elements in Chrome and send DOM, styles, accessibility data, comments, and screenshots directly to the Agent.

• 🔍 dsh-revdiff: Interactive Git diff review inside DeepSeek Harness, with structured annotations sent directly back to the current Agent session.

• 🎛️ dsh-prompt-profile: Reusable Markdown prompt profiles for DeepSeek Harness with per-turn model selection, argument substitution, and automatic state restoration.

• 🔎 ja-stopword-filter: A lightweight Python library for filtering Japanese stopwords with customizable rules for NLP preprocessing.

• ✍️ ja-ai-polish: An Agent Skill for writing natural Japanese and reducing templated AI-writing patterns while preserving facts, voice, and intent.

Projects I developed or contributed to during my previous role:

• 🏢 Japanese Company Lexicon: A high-coverage lexicon for Japanese company name recognition, developed as part of the ANLP 2020 work High Coverage Lexicon for Japanese Company Name Recognition.

• 🏷️ SeqAL: A sequence labeling active learning framework based on Flair, designed to reduce annotation effort for named entity recognition and other sequence labeling tasks.

• 🖼️ SegAL: An active learning framework for semantic segmentation, designed to support efficient data selection and annotation workflows.

• 📖 用Python动手学强化学习（Pythonで学ぶ強化学習）

Building at the intersection of:

AI Agents × Harness Engineering × Developer Tools × NLP

I like turning emerging agent ideas into small, practical tools and learning what actually works by building them.