A software engineer and AI product builder with experience developing backend systems, full-stack applications, retrieval and ranking algorithms, RAG systems, and AI-driven educational products. Has co-founded and led engineering initiatives for products serving more than 15,000 users, optimized systems at million-user scale, and conducted research in medical question-answering and real-time transcription.
Developed the DOGTOR app with more than 15,000 users, ranking number one on the App Store and Google Play in the Education category for three consecutive days with paying users. Responsible for payment flow, backend development, and data insight. Achieved product-market fit for a K-12 product with 55% retention on the second day and 25% retention on the 30th day. Trained TextCNN topic-classification models, reducing token cost by 30%. Built AI-driven question generation using domain knowledge, prompt engineering, and three-model cross-validation to generate high-quality questions and answers.
Founding Engineer — Humetric
→ Present
Developed a retrieval and ranking algorithm in MCP for HR and recruiting use cases with 2 billion users. Optimized GEO and shipped a five-locale FAQ and guides layer with server-rendered JSON-LD, after which AI-crawler traffic increased by 72.2% and the product ranked number one for relevant keywords.
Software Engineering Intern — LINE Taiwan
→ Present
Developed AI-native B2B website chatbot projects using Superpower and RAG with evaluation. Worked as a full-stack developer on the OAV, or LINE Official Account Verification, system using Go and Vue, applying clean architecture and Scrum development. Applied Terraform to manage projects running on Kubernetes using infrastructure as code.
Research & Development — Forward Alliance
→
Designed an AI real-time radio transcription pipeline to improve hospital emergency scenarios. Implemented an open-source GIS npm package and MCP server with OAuth 2. Patched user limits and deployed the open-core Mattermost project, reducing costs by more than NT$200,000. Optimized database N+1 queries for more than 30,000 users on a monorepo.
Backend Developer & Maintainer — National Taiwan University
→
Maintained the Course Selection System and Allocation Algorithm using Elasticsearch and Redis, applying layered architecture and query fallback. Applied Prisma database migrations to manage schema changes for go-live systems at million-user scale.
Research Assistant — Academia Sinica
→
Applied RAG and RLHF to improve medical question-answering chatbot response accuracy by 8%, reducing the human cost of replying to patients' questions.
Built a personal AI agent and second brain. Applied intent classification and Mem0 memory design to achieve customized long-term memory for more than 100 users.
Real Time Light Dance System
Arduino · MQTT
Developed a distributed LED choreography system for group performances using Arduino and MQTT.
Trendy Word Predictor Text Mining Final Project
Selenium · BERT · NLP · Text Mining
Crawled data from PTT using Selenium, then cleaned and labeled the data. Used BERT-Chinese_L-12_H-768_A-12 to generate word embeddings and trained a predictor capable of identifying potential trendy words from a given article. Received second place in the final project.
Education
National Taiwan University — Bachelor, Information Management