Junsu Lee
Professional Summary
Computer Science senior graduating December 2026. Deployed a Boston Dynamics Spot inspection program in a live factory, fine-tuned and physically evaluated a bimanual vision-language-action model, and connected Codex/OpenCodex to a monitored local Qwen 27B serving stack.
Education
Case Western Reserve University
Bachelor of Arts in Computer Science | GPA 3.47
Expected December 2026 | Cleveland, OH
- Coursework: Intro Artificial Intelligence, Fundamentals of Robotics, Computational Perception, Software Engineering.
- President, CWRU ACM Student Chapter (August 2026 - Present).
Relevant Experience
Advanced Manufacturing Engineer Co-op
Bendix Commercial Vehicle Systems
January 2026 - Present | Avon, OH
- Took Boston Dynamics Spot inspection from manual setup to production operation in a live factory. Configured approximately 500 inspection points, moving plant-wide collection from roughly every six months to daily or every-other-day autonomous operation.
- Shipped eight field-used tools for Orbit run-history review, live sensor diagnostics, map validation, worker signaling, and alerts. Built map and inspection-point editing workflows that cut a full update from at least one person-week to within one workday while retaining human approval.
- Built image- and log-rich alerts that cut incident discovery and response from as long as 12-24 hours to approximately 15-30 minutes; integrated Spot, GraphNav, Orbit APIs, Protocol Buffers, camera, depth, and robot-state data.
Weather AI Research Intern
xLab at Case Western Reserve University | Progressive Insurance partner
June 2025 - August 2025 | Cleveland, OH
- Owned model execution and backend delivery on a three-student R&D team. Built a Python Earth2Studio/FourCastNet pipeline on six A6000 GPUs, generating seven-day forecasts in about three minutes and publishing NetCDF4 and preview outputs to AWS S3.
Frontend Developer, AI Call Center Training System
xLab at Case Western Reserve University | Eaton partner
January 2025 - May 2025 | Cleveland, OH
- Built Next.js/TypeScript frontend components and integrated FastAPI API calls for an AI customer-simulation proof of concept designed to help call-center agents practice technical conversations with varied emotions and multilingual scenarios.
Squad Leader / IT Technician (Signaler)
Republic of Korea Army
January 2023 - July 2024 | South Korea
- Improved CCTV field installation through bench-testing and preassembly, reducing on-site installation time by approximately 70%; trained a five-person team supporting more than 400 networked devices and 34 CCTV units while troubleshooting LAN and communications equipment.
Projects
Language-Guided Arm Selection on a Bimanual SO-101 with NVIDIA GR00T N1.5
Owned the core pipeline for a custom bimanual SO-101 embodiment, from three-camera teleoperation and dataset repair to training and physical evaluation. Re-recorded approximately 360 validated episodes totaling about 238,000 frames after diagnosing timestamp corruption. Fine-tuned GR00T N1.5 on H200/L40S GPUs, cut projected training time from about 15 hours to 2.5 hours, and reached approximately 25% full-task success across roughly 20 physical trials versus 0% untuned.
Local Qwen Inference and Observability System
Operate an always-on Ubuntu/RTX 5090 server for local Qwen inference via Codex/OpenCodex, llama-swap, and vLLM; AI handled most configuration after OS installation. Switch vLLM, ComfyUI, and WhisperX with one GPU-resident service at a time; ComfyUI/WhisperX run in Docker. Services restart after reboot; Tailscale enables remote access. Monitor GPU temperature/utilization/power, vLLM throughput/requests/KV cache, and CPU/RAM/disk in Grafana.
Spot Fiducial AprilTag Plate Generator
Built a public JavaScript/SVG tool for Spot localization tags; automated SVG-to-STEP conversion through Fusion 360, cutting per-tag CAD preparation from up to two hours to roughly 10 seconds across more than 120 plates.
Data Pipelines and Visualization
Built Python/SQL ELT workflows and six tested dbt analytical models; orchestrated an Airflow/PostgreSQL environmental-data pipeline with Tableau dashboards.
Skills
Languages: Python, TypeScript/JavaScript, Java, SQL, Bash; C/C++ (coursework)
AI & ML: PyTorch, TensorFlow, OpenCV, Hugging Face, Weights & Biases, NumPy
Robotics & Physical AI: Boston Dynamics Spot, GraphNav, Orbit APIs, SO-101, GR00T N1.5, LeRobot, Foxglove
LLM Systems & GenAI: vLLM, llama-swap, Qwen, Grafana; Gemini API, RAG, function calling (Google/Kaggle labs)
Software & Data: React, Next.js, FastAPI (API integration), Pandas, dbt, Airflow, PostgreSQL
Infrastructure & Tools: Git, Docker/Podman, Linux/HPC, AWS S3, Makefile, tmux
CAD & Fabrication: Fusion 360, SolidWorks (CSWA), 3D printing
IT & Networking: LAN/switch diagnostics, CCTV, IP telephony
Certifications & Training
NVIDIA-Certified Associate: Generative AI LLMs
NVIDIA
Valid through October 2027
5-Day Generative AI Intensive (hands-on labs)
Google & Kaggle
November 2024
Deep Learning Specialization
DeepLearning.AI
March 2024
Prefer a one-page copy? Use the PDF download above.