Zichuan Xu

About

From security and database tooling to agent systems and full-stack products.

Useful agents need more than a model. They need tools, context, and a product around them.

At Initial. S, I develop KuraClaw, KuraWork, and their shared cloud services. My work connects agent execution and tool integrations with React interfaces, Go APIs, and the authentication, model routing, and usage accounting behind them.

I work across the boundaries.
From a tool call to a completed workflow.
From the backend contract to the user experience.

Experience

5 stops · 2022 → now

Initial. S Co. Ltd.

AI Agent Development Engineer · Tokyo, Japan

NowApr 2026 → now

Building agent products for everyday work: KuraWork, a desktop workspace for expert-guided tasks and business workflows, and KuraClaw, a personal assistant connecting knowledge, files, and daily tools. I contribute across the product interfaces, agent harnesses, and shared cloud services.

KuraWork

From an open-ended prompt to a configured workflow.

Helped develop a work-oriented product around the native Codex engine, with searchable Inspiration templates, domain experts, and project spaces. Templates prepare editable task drafts; projects carry default experts, skills, and connector selections that individual tasks can override. This packages reusable task setup around the engine rather than asking users to assemble it each time.

Execution connected to real work.

Integrated native tool execution, permissions, and delegated-agent conversations with the desktop UI and model services. Connected account, model, and shared-file workflows across React, Electron, and cloud APIs, bringing model selection, subscription usage, and role-controlled file sharing into the desktop workspace.

KuraClaw

Searchable personal knowledge.

Developed a memory repository with hybrid search, article relationships, and interactive knowledge graphs, alongside agent-accessible mail search and retrieval. Connected Go data services to browsing, filtering, and reading interfaces so users and agents can find and reuse information beyond the current conversation.

Daily assistance beyond chat.

Built calendar views, recurring events, reminders, and task controls, with reminders delivered through agent channels. Extended email, calendar, cloud-drive, and GitHub integrations; supported scheduled workflows and background transcription alongside the product’s photo library and file management. These capabilities connect information, actions, and follow-up in one assistant.

Shared platform

Full-stack delivery.

Developed model access, product-specific accounts and billing, and versioned shared cloud storage on AWS. Added role-based file access, quota reservations, usage settlement, and streaming recovery to support both products while preserving existing client contracts. Backed changes with regression tests and targeted end-to-end acceptance checks.

Previous stop · Rakuten 2025 Go · TypeScript · React · Electron · Codex · MCP · AWS Lambda · DynamoDB · Terraform

Rakuten

Application Engineer · Tokyo, Japan

Sep 2025 → Oct 2025

Turned two manual, ticket-driven internal workflows into real-time self-service platforms.

  • Two internal web applications in PHP/Laravel and Bootstrap 5, replacing ticket-based request handling with self-service.
  • A data automation pipeline in Python and Apache Airflow for Rakuten Card’s monthly validation process.
  • Held above 85% unit-test line coverage with zero SonarQube code smells across all deliverables.
Previous stop · Dameng 2023–25 PHP · Laravel · Python · Apache Airflow · SonarQube

Wuhan Dameng Data Co. Ltd.

Development Engineer · Shanghai, China

Jul 2023 → Feb 2025

Rendering and tooling for a next-generation database management client.

  • Built a massive-dataset rendering engine on the NatTable framework, replacing pure SWT tables.
  • Cut query display time for 100,000-row result sets from 10 minutes to 3 seconds — a 200× speedup.
  • Engineered a Universal Installer framework that decoupled environment-specific dependencies for cross-platform deployment of the whole database suite.
  • Led an embedded interactive terminal for direct SQL execution and a SQL formatter.
  • Owned the end-to-end SDLC for core modules, from requirement analysis to technical documentation.
Previous stop · NSFOCUS 2023 Java · SWT · NatTable · Eclipse RCP · SQL

NSFOCUS

Research and Development Engineer · Wuhan, China

Feb 2023 → Apr 2023
  • Gathered intelligence on fraudulent applications with Python Scrapy, and built the Django backend of a network probing system.
  • Devised encryption schemes to anonymize customer data for machine-learning training.
  • Ran technology assessments and routine maintenance for customer-facing issues.
Previous stop · Virginia Tech 2022 Python · Scrapy · Django · Cryptography

Virginia Tech

Undergraduate Research Assistant · Blacksburg, Virginia, USA

Apr 2022 → Sep 2022

Theme: a graph learning framework for system provenance-based security analysis. Advised by Prof. Peng Gao.

  • Researched graph neural networks over system provenance graphs to identify Advanced Persistent Threat behaviour.
  • Designed a graph representation learning framework in PyTorch that extracts topological features from massive, sparse system audit logs.
  • Reached 97% accuracy predicting malicious system entities while holding a low false-positive rate on large-scale security datasets.
Origin · HUST 2019 PyTorch · GNN · Provenance graphs · APT detection

Education

3 stations · 2019 → 2027

The University of Osaka

Master of Information Science and Technology · Suita, Osaka

2025 → 2027

Graduate School of Information Science and Technology. Advised by Prof. Chuan Xiao. Researching LLMs, Agents, and Benchmarks.

HUST-CSE-Experiments

Open source

A four-year archive of coursework from HUST’s School of Cyber Science and Engineering — 33+ courses spanning cryptography, software and network security, reverse engineering, compilers, operating systems, blockchain and trusted computing. The most-used thing I have published.

CAssemblySecuritySystems
View code ↗

NimGet

Open source

A lightweight download manager built in Go, with a desktop app for macOS and Windows and a CLI for Linux too. Parallel connections and crash-safe resume keep downloads moving; a persistent queue, speed limits, scheduling, and HLS support handle longer workflows. The React desktop interface and CLI share the same background service.

GoReactWailsSQLite

Skills & Stack

Full-stack development

  • TypeScript
  • React
  • Vue.js
  • Node.js
  • Go
  • Python
  • Java
  • Spring Boot
  • Django
  • Laravel
  • PostgreSQL
  • MySQL
  • Redis

AI & ML

  • PyTorch
  • Hugging Face
  • ONNX
  • Codex
  • Claude Code
  • Amazon Bedrock

Cloud & DevOps

  • AWS
  • Docker
  • Terraform
  • Linux
  • Git
  • GitHub
  • Apache Airflow

Languages

  • Chinese Native
  • English IELTS 7.5 · C1
  • Japanese JLPT N1 · B2