Software engineer with a CS degree from Taiwan Tech. I build backend and systems software — distributed task scheduling, LLM agent orchestration, embedded/BMC simulation — and do deep-learning research on cross-domain few-shot learning.

A provider-neutral agent orchestration framework. Give it a goal and it decomposes the goal into a verified task DAG, then runs each step with agents that can actually act — read and write files, run shell commands and tests, and make HTTP requests. Works with Anthropic, OpenAI, Gemini, Ollama, or any OpenAI-compatible gateway, and ships with a CLI, a Python SDK, and a React web GUI.

Python · asyncio · Pydantic · FastAPI · React · TypeScript

A distributed task scheduling and execution platform inspired by Apache Airflow and Celery. Supports the full task lifecycle, priority queues, delayed and cron-based scheduling, retries with exponential backoff, DAG workflows with a drag-and-drop visual builder, Redis sliding-window rate limiting, and real-time status push over WebSocket.

FastAPI · Celery · PostgreSQL · Redis · React · Docker Compose

A self-contained simulator of a server BMC (Baseboard Management Controller): a poll()-based C daemon that simulates sensors and fault injection, with a DMTF Redfish REST API on top. Every response is validated against the official Redfish schemas in CI, and the C code is tested with Unity under ASan/UBSan. No hardware required — just docker compose up.

C11 · Python · FastAPI · Redfish · Docker · GitHub Actions

Research on feature-level super-resolution for vision-language cross-domain few-shot learning. Upsamples ViT patch tokens from 14×14 to 28×28 with a cross-resolution consistency loss, built on cycle-consistent vision-language alignment (CC-CDFSL). Improves 5-way 5-shot accuracy on all four benchmarks (EuroSAT, CropDiseases, ISIC 2018, ChestX), +1.15% on average.

Python · PyTorch · CLIP · LoRA

The classic game: uncover every block that doesn't hide a mine.

A two-player chess game, played on the board until one side's king is captured.