Associate Software Architect & AI Innovator
I am an Associate Architect and Assistant Principal Engineer at Samsung
Electronics with over 14 years of experience in Mobile and Wearables. I specialize in building robust, performant
systems and have contributed significantly to Samsung Health for iOS (6M+ users). My work
includes seven patent applications, with one granted.
Currently, I lead the Health Sensing Modalities Team, pioneering AI-driven health sensing and on-device
intelligence.
Key innovations include stress detection algorithms for earbuds, non-invasive sweat sensors (with Caltech), and
DLMO estimation (with MIT). I also led the development of on-device foundation models for blood pressure and heart
health monitoring and conducted clinical studies on women's health.
My expertise spans Digital Health, Mobile development, System Design, and BLE. Beyond tech, I am passionate
about AI, physics, philosophy, and music, viewing software engineering as a means to solve real-world problems
and improve human well-being.
A comprehensive Java Scientific Computing Library inspired by Python's SciPy, bridging the gap in the Java ecosystem for robust signal processing. It provides high-performance, dependency-free implementations of Butterworth/Chebyshev filters, FFT, Peak Detection, and RK4 solvers—essential tools for building efficient ML pipelines on JVM and Android.
A comprehensive, full-stack contact tracing framework designed to combat pandemics like COVID-19. Utilizing a privacy-first approach, it leverages multi-modal sensing (Bluetooth, GPS, and NFC) to accurately trace close-contact candidates over a configurable period (e.g., 14 days). The system features a scalable backend architecture capable of handling nationwide data, aiming to flatten the infection curve while balancing public health necessities with individual data privacy.
A cutting-edge solution for patent prior art search that harnesses the power of Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG). By integrating the Google Patent Database, DuckDuckGo search, and advanced AI models like Google Gemini, it automates the complex process of keyword generation, query construction, and result parsing. This tool significantly streamlines the workflow for legal professionals, delivering highly relevant similarity search results with minimal manual effort.
A production-ready Chrome extension that transforms how developers and users interact with web content. InspectorAI allows you to visually select any DOM element, intelligently cleaning the HTML to minimize token usage, and instantly analyze it using top-tier AI models (OpenAI, Gemini, Anthropic, or local LLMs). Whether for reverse-engineering UI components, extracting complex table data, or translating content in-place, it supercharges web analysis without ever leaving the browser.
An essential VS Code extension for power users of AI coding assistants, ensuring fully uninterrupted workflows. NoQuota features an intelligent fallback system that automatically switches providers (e.g., from OpenAI to Claude, DeepSeek, or local Ollama models) the moment a quota limit is reached. With built-in support for a wide array of models and context-aware code completion, it guarantees that your development momentum is never stalled by API rate limits or exhaustion.
A powerful productivity suite designed for engineers to track daily work activities and generate professional status reports effortlessly. Offering both a sleek CLI and a modern Web Dashboard, FastRep utilizes AI agents (Cline, OpenAI, Gemini) to intelligently summarize your logs into weekly or monthly reports. It features a privacy-first architecture with local SQLite storage and offers over 7 customizable, live-preview templates to match any corporate reporting style.
A definitive, minimal yet complete reference implementation of the Model Context Protocol (MCP), demonstrating the future of AI-tool integration. This project provides full-stack examples of both local (stdio-based) and remote (SSE/Cloudflare Workers) server architectures. It includes a comprehensive suite of pre-built tools, prompts, and resources—ranging from creative writing to system utilities—serving as the perfect blueprint for developers building their own MCP-compliant agents and clients.
Hierarchical Masked Autoencoders for finding structure in wearable time series. A self-supervised learning framework discovering scale-sensitive structures in physiological signals for efficient on-device processing.
Apparatus and method for controlling wearable devices via strap-based tap gestures. Enables interaction without screen touch, ideal for active scenarios or gloved usage.