02 / Experience

Five roles across research and production.

AI agents on edge hardware, full-stack web with Laravel + Airflow, database rendering engines, security R&D, and graph learning for system provenance — selected highlights below.

2022 → Present · 3 countries · 5 organisations

Apr 2026 — Present

AI Agent Development Engineer Intern

Initial S Co., Ltd.·Tokyo, Japan

Building Kuraclaw — an AI agent designed to run efficiently on NAS-class hardware rather than data-centre GPUs. The work spans the edge between model serving and orchestration: how do you keep an agent useful when every token has a measurable cost in watts and latency?

Optimised LLM inference and tool execution for NAS-class hardware, trading off context window, batching strategy, and tool-call granularity
Designed efficient agentic workflows that reduce end-to-end latency and resource overhead under typical user prompts
Engineered backend components and tool-calling mechanisms; profiled hotspots and rewrote the worst offenders

LLM Inference AI Agents Tool Calling Edge Hardware

Sept 2025 — Oct 2025

Application Engineer Intern

Rakuten Group, Inc.·Tokyo, Japan

A short, intense full-stack rotation. Shipped two internal web applications end-to-end and stood up a Python data pipeline that replaced a brittle hand-run script.

Built two internal web applications in PHP / Laravel + Bootstrap 5, owning schema, controllers, views, and deployment
Designed and implemented a data automation pipeline in Python and Apache Airflow, scheduled and monitored from a single DAG
Held the test bar at over 85% unit-test line coverage on the application code

PHP · Laravel Airflow Python Testing

Jul 2023 — Feb 2025

Development Engineer

Shanghai Dameng Database Co., Ltd.·Shanghai, China

Worked on the developer-facing surface of a commercial RDBMS — the parts users actually see when they open the client. A lot of the gain here came from one rewrite of the query-result rendering engine.

Architected a new rendering engine for query result display, reducing time-to-first-row from ~10 min down to 3 s on large result sets
Engineered a Universal Installer framework — a single declarative spec drives installer behaviour across product lines
Led the SQL terminal and SQL formatter implementation, including syntax-aware indentation, multi-statement parsing, and inline diagnostics

Java SWT · NatTable SQL Performance

Feb 2023 — Apr 2023

Research & Development Intern

NSFOCUS Technologies Group Co., Ltd.·Wuhan, China

A short R&D placement on the security side. Wrote crawlers and the backend that ingested their output, plus a small encryption layer used to anonymise sensitive samples before sharing them across teams.

Developed security-oriented data acquisition tools in Python / Scrapy
Built a Django backend to serve and query the collected dataset
Devised encryption / anonymisation routines for sharing data across internal teams

Python · Scrapy Django Security

Apr 2022 — Sept 2022

Research Intern

Virginia Tech·Prof. Peng Gao's group·Blacksburg, VA

My first deep dive into research. Worked on a graph learning framework for system-provenance-based security — modelling system call traces as graphs and using GNNs to flag malicious entities inside them.

Researched graph neural network architectures for advanced persistent threat (APT) detection on provenance graphs
Reached 97% accuracy on the malicious-entity prediction benchmark used by the group — state of the art at the time
Contributed to dataset preprocessing, baseline reimplementations, and ablation experiments

PyTorch GNN APT Detection Provenance

Back · 01

Education

Home