Full-Stack & Robotics Software Engineer
Jun Hur
I build software that keeps robot fleets running in the real world.
For 5+ years I have built production systems across robotics, fintech, and healthcare. I am applying to US CS master's programs.
01. Research Interests
My work sits where software meets the physical world. Building and operating robot fleets and data pipelines in production led me to three research interests in software engineering for cyber-physical systems:
01 Fault Diagnosis Across System Layers
Fleet failures span the control frontend, backend, message broker, robot software, and site network. I am interested in observability and fault-localization techniques that trace causes across these layers — a problem I faced daily while operating 30+ robots in production.
02 Failure Prediction from Operational Telemetry
At DOGU Robotics I found that anomalies in MQTT message frequency repeatedly preceded customer-reported failures. I am interested in how far such early-warning indicators generalize across systems.
03 Simulation-Based Validation Before Deployment
I build validation workflows in NVIDIA Isaac Sim / Isaac Lab — today for automation retrofits on heavy construction equipment, earlier for autonomous mobile robots — and wrote a from-scratch MAPF simulator for pre-deployment fleet sizing. I am interested in scenario generation and in quantifying what simulation guarantees about field reliability.
02. About
I am a senior software engineer at Xpanner, building control software and simulation-based validation workflows for automation retrofits on heavy construction equipment. Before that I built the fleet management system for autonomous robots on construction sites at GoLe Robotics, was the sole developer behind 200,000+ auto-generated corporate carbon-risk reports at the fintech startup Niflers, and led web development for robot fleet control at DOGU Robotics — 30+ robots across 10+ sites.
Alongside production work I keep building fundamentals and measuring the results: re-implementing the Transformer and xLSTM from the original papers, and writing a multi-agent path-finding (MAPF) simulator from scratch in TypeScript. My undergraduate research, published at KCC 2024, measured a 70–96% reduction in nursing-documentation time on simulated records.
I am applying to US master's programs in computer science to study the reliability of cyber-physical systems — fault diagnosis, failure prediction, and simulation-based validation — and intend to continue to a PhD.
03. Skills
04. Experience
Present
- Developing the control and management software that connects machine-level signals to operator consoles and cloud systems, and keeps that link reliable in real construction environments.
- Building simulation-based development and validation workflows with NVIDIA Isaac Sim and Isaac Lab, so control logic can be validated before it is deployed to physical machines.
- Improving and stabilizing the equipment-operations platform and its AWS infrastructure for telemetry ingestion and remote equipment management.
2026.08
- Sole developer of the fleet management system (FMS) for autonomous delivery robots — web operations console, companion mobile app, and browser-based driving simulator.
- Implemented real-time control and monitoring over the MQTT-based VDA5050 standard, with WebSocket live feeds that throttle rendering and reconnect automatically for unattended sites.
- Built the visual scenario editor (routes drawn over SLAM maps, compiled into VDA5050 orders), cargo tracking with auto-captured handover photos, and elevator-integrated multi-floor delivery.
- Built an NVIDIA Isaac Sim environment — CAD-imported URDF robot, unmodified ROS2 nodes, Nav2 navigation — and validated a complete multi-floor delivery sequence in simulation before on-site runs.
- Began building Isaac Lab reinforcement-learning pipelines with domain randomization toward sim-to-real transfer.
2025.10
- Owned the full stack of three fintech products — CRISK, Credit Imbalance Tracker, and Credivalue — from database design to deployment.
- Built the reporting system behind 200,000+ auto-generated carbon-risk reports and refactored its 20,000+ line monolithic script, cutting generation time from 10 minutes to 2.
- Designed the MySQL schemas and daily ingestion pipeline for KED and FnGuide financial data, with parallel loads on AWS ECS + Step Functions.
- Diagnosed and resolved two production incidents — a data-integrity failure during 20-process parallel loading and a framework upgrade that broke every API.
- Prototyped an LLM-based question-answering interface for querying reports in natural language.
2023.05
- Led web development of the fleet-control platform managing 30+ robots across 10+ sites — hospitals, factories, and outdoor environments — for clients including Hyundai, SK hynix, and GS EPS.
- Implemented WebRTC video streaming, MQTT-based robot control and monitoring, and Grafana dashboards; migrated the frontend from Vue.js to React and standardized components across 10+ site-specific deployments.
- Debugged incidents end to end across the control frontend, backend, MQTT broker, and robot ROS layers, teaching myself each layer beyond my original frontend role.
- Found that anomalies in MQTT message frequency often preceded customer-reported robot failures, and built monitoring on that signal with no extra instrumentation.
- Shipped branching diagnostic scripts that field engineers used to localize faults by elimination — battery, then network, then software.
2021.11
- Developed and maintained ERP web services for Incheon City Gas.
2021.04
- Converted nuclear power plant calculation logic into a C++ digital twin in collaboration with K-water (Korea Water Resources Corporation).
05. Projects
Web console, mobile app, and driving simulator for autonomous delivery-robot fleets — real-time VDA5050 control over MQTT/WebSocket, scenario editor, cargo tracking, multi-floor delivery.
NVIDIA Isaac Sim environment with a CAD-imported AMR, ROS2/Nav2 navigation in RViz, and a simulated elevator — validates full multi-floor delivery sequences before on-site runs. Demo videos inside.
Corporate carbon-risk analysis platform — automated financial-data ingestion, MySQL storage, and FastAPI-based report generation.
Financial-health dashboards (CIT Score, Industrial Risk, Early Warning Signal) for banks and asset appraisers, built end to end.
Present-value valuation platform — Excel uploads trigger asynchronous processing on AWS Lambda + Step Functions + SQS, with full CI/CD.
Real-time monitoring and control for autonomous robots in hospitals, factories, and corporate environments — WebRTC streaming and MQTT control.
Touch UI for service robots' front displays — information kiosk, indoor navigation, and live video calls via the Zoom SDK.
C++ digital-twin simulation of nuclear plant thermal-hydraulic calculation logic — validated together with domain engineers.
Speech-to-text platform that turns clinical conversations into structured records — backend services and the AI-integration workflow.
Selected Open-Source Projects
Repositories on github.com/hurjun. Each documents how to reproduce its results; the numbers below are measurements recorded in each README.
3D multi-robot fleet simulation on multi-floor sites with a MAPF engine written from scratch in TypeScript (windowed cooperative A*, Conflict-Based Search). Zero collisions across 66 headless benchmark runs / 21,000+ ticks.
Edge-to-cloud pipeline for site-safety alerts: YOLOv8 edge clients (stock weights, PPE model pluggable) send events to a FastAPI hub, which fans them out over WebSocket with backpressure. 33 tests; detection accuracy not evaluated.
Attention Is All You Need re-implemented from primitives in PyTorch. Reaches 98.2% exact sequence match on a synthetic reversal task (CPU, ~90 s). In a separate 1,200-step ablation, removing positional encoding drops exact match from 91.2% to 0%.
sLSTM and mLSTM cells (Beck et al., NeurIPS 2024) implemented from scratch in PyTorch. On a 64-step recall task, xLSTM reaches 100% accuracy while a vanilla LSTM stays at chance (≈6.25%), even with up to 2× the parameters.
Speculative decoding with distribution-preserving acceptance sampling, implemented from scratch. Up to 1.80× lossless speedup on Apple Silicon (DistilGPT-2 drafting, GPT-2-large verifying), with output token-identical to greedy decoding.
Voice-first nursing documentation: an AI speaker asks assessment questions (TTS), transcribes the answers (STT), and normalizes them into structured records. Implementation of the KCC 2024 paper (3rd of 4 authors): documentation time −70% on first use, −96% on repeated use, on simulated records.
Leakage-controlled benchmark of 9 forecasting models, from ARIMA to Transformer, on 1962–1992 → 1993–2000 stock data. Prophet ranks first (RMSE 160), with an analysis of why the deep-learning models underperform.
06. Education & Publication
Third of four authors: system implementation and evaluation design. Undergraduate thesis project (team of three, advised by Prof. Jihun Park). In a five-trial timing evaluation on simulated patient records, the assisted workflow cut documentation time by 70% on first use and 96% on repeated use.
Completed the degree while working full-time as a software engineer from 2021 onward.
For Reacord, a real-time lecture-transcription app built and released for deaf and hard-of-hearing classmates.