BongKyu Lee
Objective-C · Swift
Robotics · Cloud systems
Korea · US · Japan
M.S.
About
I’m a software engineer and independent researcher. Over roughly fifteen years I’ve shipped production software across Korea, the United States, and Japan — moving from low-level systems, embedded controllers, and native apps toward enterprise platforms and, more recently, applied AI and robotics.
My current work sits at the intersection of cloud and industrial-systems engineering and LLM integration. Alongside it I run Kyuling Company, through which I build B2B SaaS for the Japanese market. The through-line has stayed the same throughout: understanding a system end to end — from the operating system underneath to the product someone actually uses — and building the part that was missing.
I’ve kept one foot in research the whole way. My background is in systems software (M.S., Chung-Ang University), and I’m currently working on reinforcement learning for whole-body humanoid locomotion, with results published and an arXiv submission in progress. Picking up an unfamiliar language, stack, or field and becoming useful quickly has been less a phase than a working habit.
I care about work that holds up — systems that are correct, legible, and still make sense a few years later — and I’m most engaged when a problem makes me learn something I didn’t know yesterday. That part hasn’t changed with time; if anything, the problems have only gotten more interesting.
Publication
M.S. thesis · 2018Social Networks Implementing the Internet of Things through Controlled Natural-Language Conversations
자연어 대화를 이용한 사물인터넷 제어가 가능한 소셜 네트워크 구현
This thesis treats IoT devices as first-class members of a social network: each device is given its own account and profile and participates in the same feed as people. A small-scale social network was built in which short, controlled natural-language posts are parsed to decide which device should respond and what action it should take. The system issues the command, reads the device’s result back, and replies in the same thread — demonstrating that everyday objects can be understood, controlled, and have their state reported entirely through social-network conversation.
국문 초록 펼치기
사람과 사물이 인터넷에 쉽게 연결될 수 있는 사물인터넷 분야가 연구되고 발전하고 있다. 그 중에서도 사물 간의 상호성에 대한 연결, 관계에 대한 연구가 관심을 얻기 시작하였다. 추가적으로 사람이 사물과 자연어를 통해 대화를 할 것으로도 내다보게 되었다. 따라서 사물인터넷 환경에서 사람 및 사물, 사물 및 사물의 상호 연결 관계를 만들 수 있는 환경으로 소셜 네트워크를 이용하는 방법을 생각하였다. 본 논문에서는 소규모의 소셜 네트워크 환경을 구축하고 사람과 사물이 기계적인 대화를 할 수 있도록 실험하였다. 소셜 네트워크상에서 사물 또한 사람처럼 대화할 수 있을 수준의 개체로 인식할 수 있게 고유의 계정과 관리 시스템을 구축하였고, 장치의 세부 정보를 프로파일로써 저장하여 관리할 수 있게 하였다. 대화를 위해 소셜 네트워크에 글을 작성하면 해당 글을 읽어들여 어떠한 사물과 대화를 진행할지 결정하고, 어떠한 정보를 받아들여서 실행해야 하는지를 판단한다. 그에 따라 추론된 결과와 제어할 장치 정보를 이용하여 장치에게 명령을 전달하고, 장치의 실행 결과를 받아와서 다시 답변하는 형태로 구현을 진행하였다. 대화 수준은 간단한 기계적 수준의 대화를 진행하였다. 이를 통해 소셜 네트워크상에서 사물이 사람과 같이 대화를 알아듣고, 그에 따라 제어되고, 그 결과를 알려줌으로써 대화를 통해 사물을 제어할 수 있다.
Consulting & Applied AI
Accenture · 2024–presentDocument Sanitization & AI-Ready Data Pipelines
Cleaning, sanitizing, structuring, and vectorizing large enterprise document sets so they can be consumed safely by LLM-based systems and retrieval — the groundwork that makes downstream AI usable on real corporate data.
Domain-Specialized AI Assistants (RAG)
Assistants that pair retrieval-augmented generation with domain-tuned models, so non-specialists can work through expert material with AI support. Prototypes have spanned customer-facing sales assistance and regulated domains such as clinical safety review (pharmacovigilance).
Multi-Agent, Voice-Controlled Robotics
A voice-controlled robotics demo built on a multi-agent orchestration framework: a spoken natural-language instruction is parsed, the target object is identified by camera, a motion command is generated through ROS, and executed on an industrial robot arm. Presented at an international robotics exhibition in 2025.
Hands-Free Voice Data Entry for Controlled Environments
A hands-free, voice-driven data-entry tool designed to work inside tightly controlled clean-room environments where touch input isn’t practical, delivered together with companion wearable hardware.
Korean Localization & Review
Korean-language localization and quality review supporting product rollouts in the Korean market.
Portfolio
Product engineering · 2011–2023Ship Alarm-Monitoring & Control System — OPS / FCU
At JRCS (Japan), developed software for a marine Alarm Monitoring & Control System (AMS) — the OPS (Operator Station) and FCU (Field Control Unit) components — across the full lifecycle: requirements, basic & detailed design, development, test, and maintenance.
- OPS (Operator Station) — Windows-side operator software in C / C++ / C# on .NET Framework, backed by PostgreSQL.
- FCU (Field Control Unit) — RTOS firmware in C / C++ for field-level control.
- Built per-shipyard custom features for individual vessel-construction orders, plus ongoing program maintenance.
- Implemented a range of shipyard-specific special functions across OPS and FCU, and resolved numerous software non-conformances.
- Created gateway solutions to interoperate with other vendors’ hardware and services.
Solution Development Consultant
Consulting for public institutions and companies adopting IT solutions, covering both new builds and redevelopment.
- Advised on planned services and scope.
- Drafted appeal documents and proposal-request (RFP) forms on the client’s behalf.
- Supervised interim and final deliverables, reviewing submitted results together.
Embedded Controllers on Raspberry Pi
Redeveloped mechanical-facility controllers and their internal programs, plus smartphone monitoring apps.
- New hardware controllers in two form factors: Pi 3 B+ connected, and Pi Compute Module 3 socketed.
- Control system built on open-source Qt; drivers written for the additional devices.
- Led iOS monitoring app; collaborated on Android and Windows monitoring; Bluetooth link between controllers and monitors.
My Own Interpreter
Built an interpreter that parses and executes source code, written to study compiler and interpreter internals.
- Deliberately minimal spec: only
doublevariables, one-dimensional arrays, andif/forcontrol flow. - Functions supported; execution optimization intentionally out of scope.
“AlunePOS” POS Solution
A Windows POS solution running across many POS devices, with a back-office server POS for sales and inventory. Deployed across dozens of stores in California and Nevada.
- Fingerprint recognition for attendance and login.
- License-management site plus a license-check API and program modules.
- Device-control modules covering most US-market peripherals.
- Operated a VMware vSphere private-cloud server.
IoT over Controlled-Language Conversations on Social Networks
Master’s thesis: devices live on social networks under the same accounts as people, coordinating through posts.
- Following / followed devices read posts to learn whether action is needed.
- When a device must act, it is confirmed through the post itself.
- First implemented for lighting (fluorescent-lamp) devices.
OS Memory-Management Research
Sparked by an IEEE article on memory: could equipment that must run non-stop keep using its memory while skipping physically broken cells? Study designed around the Linux kernel’s memory subsystem.
- Found no viable path once I learned how DDR memory actually fails.
- Could not build an environment to simulate and debug the experiment.
- Blocked at the design stage; the project was stopped.
Operating-System Study
Implemented core OS components from coursework, by hand, and shared them as study material.
- A working shell.
- A buddy allocator for memory management.
- A FAT filesystem exercise.
Arduino Tutorial Project
Simple, follow-along Arduino projects for people curious about making, shared on the Makewith site (since removed in a renewal).
Class-Assistance Program
Classroom software that mirrors the teacher’s example screen onto students’ PCs.
- Teachers monitor student screens in real time.
- Remote support for students who need help.
Restaurant Kiosk
An entrance kiosk that takes orders and relays them to the kitchen.
- Kitchen POS receives the orders.
- A display screen shows order information.
Embedded Controllers on Android
A controller for a machine that detects defective rice grains.
- Android NDK + Cocos2d-x used together to implement and run Linux-based device drivers.
- Driver for the grain-inspection signal reader.
- Control drivers for set-values and various switch devices.
QR Service (E-Health) iOS App
“QR Services,” an iOS app behind a patent application that encrypts hospital and pharmacy prescription data into QR codes.
- Users scan the QR code to see the type and quantity of each prescribed drug.
- Reminders for the next hospital visit and for dosing times.