// Welcome to my space

Hello, I'm Jinu Choi, Researcher.

Building accessible systems that bring real change to people's lives. HCI Researcher at Letsur.

Download CV
My Services
📸 Momentia AI-powered photo archiving & memory management
🎵 SoundVault Personal music streaming & library organizer
📄 CV Hype Machine Over-the-top resume compliment generator

Interested in trying any of these? Contact me to get access.

Building accessible systems that matter

Jinu Choi

Hi, I'm Jinu Choi, an HCI Researcher at Letsur. I'm deeply interested in system design, particularly mobile computing and user interfaces.

Previously, I researched human-in-the-loop accessibility systems for visually impaired users at KAIST NCLab.

I dive deep into new problems and prefer learning new technologies for research rather than relying on existing ones. I love bringing real change to people's lives through my work.

HCI Researcher Letsur
2025.04 ~ Present
M.S. Computer Science KAIST NCLab
2022.09 ~ 2024.08
B.S. Computer Science KAIST
2017.03 ~ 2022.08
Languages Python, Java/Kotlin, JavaScript, C/C++
Mobile & Frontend Android (Jetpack), AOSP, React
XR & Graphics Unity, AR Foundation, Oculus
AI/ML TensorFlow/TFLite, PyTorch, Scikit-learn

Things I'm thinking about

Sharing thoughts on technology, design, and the future. Hoping these ideas will evolve into projects someday.

01

AI Agent Memory System

Instead of storing answers, store functions. Extract knowledge inductively from conversations— preserving context like who, when, and under what conditions. Rule-based reasoning over value-based retrieval.

AI Memory Knowledge Graph
02

Spark System

Capture fleeting ideas in 2 seconds, auto-link to existing notes, and build meta-cognition over time. Making Claude an AI that truly knows me through thought pattern analysis.

PKM Meta-cognition Obsidian
03

Private Music Archive

A private SoundCloud for music clubs—stream, search by lyrics or credits, visualize collaboration networks. Digitize years of CDs gathering dust into a searchable archive.

SaaS Music Multi-tenant
04

AI Crystallizer

Just talk naturally—the AI gradually understands you in the background. It collects your scattered utterances, crystallizes implicit thoughts you can't quite articulate, and proactively connects ideas across domains.

AI Personalization PKM
05

N-Split

Graph-algorithm-powered group expense settler that minimizes the number of transfers using Minimum Cash Flow and Cycle Detection. Native Korean payment integration for one-click settlement.

Fintech Algorithm Graph
06

Agora

Multiple LLMs take org-chart personas—CTO, designer, junior dev—and hold structured debates on real decisions. Applies Jury Learning to intentionally surface minority opinions that majority voting would silence.

Multi-Agent Simulation Decision Making
07

Dream Pipeline

Inspired by REM sleep—an offline batch pipeline that consolidates AI memories. Discovers cross-memory patterns, creates implicit links between unrelated knowledge, and compresses old details into higher-level abstractions.

AI Memory Architecture
08

AI Memory Pipeline Principles

An anti-complexity, schema-first design philosophy for AI agent memory. Three-stage Write–Index–Read pipeline where schema quality and query decomposition beat complex retrieval algorithms every time.

AI Architecture Design Principles
09

Typed Memory System

Memory nodes typed into Knowledge, Person, and Event with distinct schemas. Events act as ambient context injected into every query automatically. An Alias Registry expands domain jargon transparently at search time.

AI Type System Search
10

CPS Agent Architecture

Dual-channel pattern for enterprise AI agents: Expert Channel treats humans as "slow tools" via A2H protocol, User Channel handles Q&A. A Messenger Layer absorbs multi-turn and multi-channel complexity.

Agent Enterprise Architecture

Projects

06 Projects
2024

Chaperone

A privacy-preserving system for visually impaired users receiving remote assistance. Modified AOSP to filter sensitive information at the framework level before sharing with helpers.

AOSP Android Privacy
Video
2023

Hivemind Field Trial

Field trial research on instant social control of shared devices in public spaces. Deployed in classrooms and cafeterias to analyze user behavior. Published in IMWUT.

OpenHAB Flask IoT
Paper
2022

Sidekick AI

A context-aware system that monitors driver status during voice chatbot conversations. Uses T Map SDK to predict road conditions and pauses conversations in high-attention zones.

Android T Map SDK Context-aware
Video
2022

ButtonTap

Wearable interaction using shirt buttons with embedded microphones to classify object contact sounds. Achieved high accuracy object recognition using SVM-based algorithms.

Wearable SVM IoT
Video
2021

AR/VR Vocational Training

AR/VR application for students with developmental disabilities to practice car-sharing vehicle inspection. Developed tablet AR and Oculus VR simultaneously using Unity and AR Foundation.

Unity AR Foundation Oculus
Details
2019

Crosswalk Navigation for VIP

App that recognizes crosswalks via camera and guides walking direction by tracking user position. Designed intuitive vibration patterns and low-vision friendly UI. (Patent filed)

Android Computer Vision Accessibility
Video Patent

Writing

AI, 제품, 그리고 생각의 구조에 대한 글.

Comments