Yu-WeiChang

1. 01● active

   UCVLA: User-Conditioned Vision-Language-Action Models

   O(1) personalization for robot foundation models via per-user bias vectors.

   advisor Prof. Jonathan Kaolab UCLA Neural Engineering and Computation Labwhen nov '25 → present

   • Validated per-user embedding divergence, measured by cosine similarity, by finetuning RDT-2 FM with frozen-backbone per-user bias vectors $b_u$.
   • Architecting $\mathcal{O}(1)$ personalization across $\pi_0$, RDT-2, and Diffusion Policy by injecting $b_u$ at conditioning points, outperforming $\mathcal{O}(N)$ LoRA.
   • Developing zarr-to-LeRobot/WebDataset pipelines and collecting 200-trial/user data via handheld UMI hardware to scale across 8–10 participants.

   • VLA
   • RDT-2
   • π₀
   • Diffusion Policy
   • PyTorch
   • UMI
   • LeRobot
   • WebDataset
2. 02● active

   EasyReach: Adaptive Robotic Arm Assistance for Wheelchair Users

   Prompt-driven preference extraction for accessible robot behavior without retraining.

   advisor Prof. Yang Zhanglab UCLA HiLabwhen sep '24 → present

   • Developing a prompt engineering framework that analyzes user videos to extract preferences for accessible robot behavior without retraining.
   • Deployed a Vicon motion-capture pipeline with <5 mm tracking precision, achieving real-world object retrieval on the physical robotic arm.
   • Prototyped kinematic feasibility in VR (Unity, Meta Quest 2), running 500+ reachability scenarios and reducing physical testing time by ~80%.

   • Vicon
   • Unity
   • Meta Quest 2
   • ROS 2
   • MoveIt
   • UR5e
   • VLM
3. 03○ completed

   VisCoder Bench: Do Coders Need Eyes?

   Benchmarking visual grounding for multimodal coding agents on 50 executable CV tasks.

   advisor Prof. Achuta Kadambilab UCLA Visual Machines Groupwhen sep '25 → nov '25

   • Developed VisCoder Bench, a standardized suite of 50 executable CV tasks (Geometric, Generative, Classical) to evaluate multimodal agents.
   • Dockerized evaluation infrastructure on A100 GPUs, integrating 10+ CV pipelines with PSNR/mAP and VLM-Judge (87% human alignment).
   • Analyzed 900+ trajectories, showing visual grounding improves geometric task success by 4.3× — exposing critical gaps in LLM reasoning.

   • VLM
   • Docker
   • A100
   • OpenCV
   • PyTorch
   • Multimodal Agents
4. 04○ completed

   Open Source USB HID Programmer for 8051 & AVR

   Driverless cross-platform flashing toolchain shipped to 200+ embedded students.

   at National Tsing Hua Universitywhen sep '22 → jun '23

   • Engineered a driverless USB HID firmware and custom packet protocol, optimizing flash write throughput to reduce programming time by 80%.
   • Deployed a cross-platform toolchain (C++/C#) adopted as production tooling for 3 undergraduate embedded systems courses (200+ students).

   • C++
   • C#
   • USB HID
   • 8051
   • AVR
   • Embedded

1. apr '25 → sep '25

   Taichung, Taiwan

   Software Engineer (Contract)Shin-Da Information Co., Ltd.

   • Built a Python serverless backend on AWS (Lambda, API Gateway), optimizing transaction latency by 35% to support 50+ enterprise clients in Q1.
   • Engineered an automated ETL data validation pipeline, which accelerated processing throughput by 40% and eliminated manual entry errors.
   • Led a 6-week initiative to architect and deploy a fault-tolerant automation platform, driving daily operations for 50+ staff members.

   • Python
   • AWS Lambda
   • API Gateway
   • ETL
   • PostgreSQL
2. mar '24 → aug '24

   Taipei, Taiwan

   Software Engineer (Contract)Marker X Biomedical Co., Ltd.

   • Architected a centralized PostgreSQL data lake unifying mass spectrometry and experimental datasets, replacing fragmented legacy storage.
   • Developed high-throughput analysis pipelines that reduced data processing latency by 90%, accelerating research iterations from days to minutes.
   • Scaled backend infrastructure to manage 10,000+ sensitive records with 100% data integrity, ensuring high availability for 30+ scientists.

   • PostgreSQL
   • Python
   • Mass Spectrometry
   • Backend
3. jul '23 → dec '23

   Hsinchu, Taiwan

   Research Development & Software Engineer InternIndustrial Technology Research Institute (ITRI)

   • Engineered a RAG-based memory system and fine-tuned PyTorch NLP models for elder-care robots, achieving 90% intent recognition.
   • Deployed a multi-modal AI agent platform with LangChain and DALL·E 3, automating generative AI workflows for 200+ staff.
   • Designed a high-concurrency Go backend for a national-scale distributed DB, securing record exchange for Taiwan's National Health Insurance.

   • Go
   • PyTorch
   • LangChain
   • RAG
   • Distributed Systems

1. sep '24 → expected jun 2026

   gpa 3.88 / 4.0

   M.S. — Electrical and Computer EngineeringUniversity of California, Los Angeles

   • Graduate Teaching Assistant — Neural Networks & Deep Learning (ECE C147/C247).
   • Manage a 500-student course: weekly technical mentorship, discussion modules, exam design, project evaluation.
2. sep '20 → jun '24

   gpa 4.04 / 4.3

   B.S. — Electrical EngineeringNational Tsing Hua University

I'm wrapping up my MS in mid-2026 and looking for robotics SWE / MLE / SWE roles. Best way to reach me is email.