Looking for a job

John Chen

@johnchen

Machine learning engineer at HiPower who developed a vision analytics pipeline matching 90B-model output with 80% lower GPU memory.

Taiwan

Message

At HiPower, I led an open-ended proof of concept for AIoT traffic and public-safety vision analytics. I designed a pipeline that replaced per-site YOLO retraining with structured decoding into an 18-field JSON schema.

I also re-engineered the system from dual NVIDIA H100 GPUs to a single RTX 5090 using a three-model subagent pipeline, matching the 90B VLM output with 80% lower GPU memory. I aligned outputs against a 60-case automated test suite.

On TechPulse, my solo-designed iOS project, I built offline learning features with Apple Foundation Models, including article summaries, term explanations, and quizzes. I also created a knowledge graph connecting 68 concepts to quizzes and user-configurable resources.

Experience

Work history, roles, and key accomplishments

Machine Learning Engineer

HiPower

Mar 2026 - Apr 2026 (1 month)

Led an open-ended proof of concept from an undefined problem space to a production pipeline, defining the technical direction for AIoT traffic and public-safety vision analytics. Designed end-to-end architecture on dual NVIDIA H100 (90B VLM) and re-engineered for a single RTX 5090 via a 3-model subagent pipeline of 11B VLMs, matching 90B output at 80% lower GPU memory.

VLM LLM Inference Structured Decoding Prompt Engineering Multi GPU Distributed Deployment fastAPI Ollama

Education

Degrees, certifications, and relevant coursework

Tamkang University

Master of Science, Computer Science and Information Engineering

2021 -

Activities and societies: Teaching Assistant, Introduction to Artificial Intelligence (Feb – Jun 2022)

Pursued a Master of Science in Computer Science and Information Engineering, with expected completion in January 2026. Served as a Teaching Assistant for Introduction to Artificial Intelligence from February to June 2022.