AI engineer · Taichung, Taiwan
I build agent systems, then measure whether they actually work.
AI engineer, backend by training, Go by default. I build agent systems and run them: taidle, my own knowledge-base agent engine, answers customer questions on three client websites today, and every reply shows the route it took to get there. Underneath that sits ML serving work: architecture chosen against a real CPU or GPU budget, deployment to cloud GPUs and Jetson edge devices, and OpenTelemetry, Prometheus and Grafana on all of it so the numbers are real.
M.S. in Computer Science, National Chung Hsing University. Two open-source research projects on ML serving — one benchmarks three architectures under CPU limits, the other implements the second study's batching proxy in Go — and a blog on what the benchmarks leave out. Open to opportunities.
Research
Batch2go: isolating batching effects by placement
Treating batching placement as an isolable architectural variable, with a single-flight Go proxy that decomposes batch cycle time into compute and transport.
A reproducible evaluation framework for ML serving architectures under CPU constraints
A controlled comparison of monolithic, microservices and Triton serving on CPU-constrained hardware. The finding is a three-way trade-off, not a ranking.