Hello, I am Yu-Yun
Software Engineer, Simulation Engineer & AI Engineer
Würzburg
About
With over 11 years of experience in computational engineering, I specialize in the unique intersection of mechanical simulation, software development, and artificial intelligence. My journey began with a foundation in mechanical engineering and evolved into developing sophisticated simulation tools and AI systems that solve complex engineering challenges.
At Wölfel Engineering, I've pioneered the development of Casimir/Automotive, a comprehensive Python/TCL scripting framework that integrates with industry-leading tools like Abaqus, LS-Dyna, and HyperWorks. My work spans from finite element analysis and seating comfort engineering to cutting-edge AI development, including the Casimir AI Assistant—an agentic AI system with RAG capabilities that runs efficiently on just 6GB VRAM.
I'm passionate about bridging the gap between traditional engineering disciplines and modern AI technologies, creating tools that enhance productivity and unlock new possibilities in computational engineering. My diverse background enables me to approach problems from multiple perspectives, whether it's optimizing wind turbine simulations or fine-tuning large language models for specialized engineering applications.
Experience
Project Engineer & AI Engineer
Wölfel Beratende Ingenieure GmbH + Co. KG
- Seating comfort engineering including FEA simulation, ride comfort analysis, and experimental modal analysis
- Lead developer of Casimir/Automotive: Python/TCL scripting framework for Abaqus, LS-Dyna, HyperWorks, and Simpack integration
- Simulation engineering for wind turbines with OpenFAST customization and optimization
- AI development: Created Casimir AI Assistant using Llama 3.1 with agentic AI architecture and RAG implementation
- LLM fine-tuning and optimization for engineering-specific applications running on limited hardware (6GB VRAM)
Development Engineer - Product Development
Adaptronics International GmbH
- Product development and engineering design
- Technical analysis and prototyping
- Collaboration with cross-functional teams on product innovation
Application Engineer - Pump Selection
KSB
- Technical consultation for pump selection and system design
- Application engineering and customer support
- Performance analysis and optimization recommendations
Skills & Education
Technical
Tools
Languages
Education
MSc Computational Engineering
Ruhr-Universität Bochum
BSc Mechanical Engineering
National Taiwan University of Science and Technology
Projects
Featured Projects
Casimir AI Assistant
An intelligent agentic AI system designed for engineering support, featuring advanced RAG (Retrieval-Augmented Generation) capabilities and multi-modal understanding.
- Answers complex engineering questions using domain-specific knowledge base
- Translates natural language queries into technical commands
- Reads and interprets error screenshots for troubleshooting
- Escalates complex issues to human support when needed
- Optimized to run on only 6GB VRAM, making AI accessible on standard hardware
Full Automation of Seating Comfort Simulation Pre-Processing
Automated preprocessing pipeline for seating comfort FEA simulations, streamlining the workflow from posture definition to ready-to-run finite element models.
- Only a posture definition file exported from RAMSIS is required
- Automated manikin model positioning
- Automated boundary conditions
- Ready-to-run FE model
Automation of Aircraft Damage Simulation Post-Processing
Automated post-processing workflow for aircraft damage analysis, enabling seamless data exchange between simulation team members.
- Automated post-processing for different FE simulations
- Outputs exchanged between team members for succeeding simulation
Design Load Case Simulations with OpenFAST for Machine Learning
Customized OpenFAST simulation framework for generating wind turbine design load cases, creating training datasets for machine learning applications.
- OpenFAST was customized for special outputs
- Automated Design Load Case Generation
- Wind turbine models with/without structure/control failures
Simulation and Optimization for Parts of Vehicle Powertrain
Comprehensive simulation and optimization project for critical powertrain components, including drive chain, hydraulic systems, and timing mechanisms.
- Simulation of motor chain drive
- Simulation of hydraulic connecting rods for combustion engine
- Simulation of crankshaft
- Timing chain tensioner optimization
Ride Comfort Analysis for a Subcompact Car
Complete vibration and comfort analysis combining experimental measurements with computational models to assess and improve vehicle seating comfort.
- Road test measurements
- Seat modal analysis
- Transfer function analysis for dynamic seating comfort
Simulation and Optimization of an Adaptive Damper for Drivetrain
The adaptive damper's eigenfrequencies become higher with rotational speed because of geometrical stiffening
- The multi-flexible bodies were mathematically modeled and programmed in Matlab
- Multi-objectively optimized for limited drivetrain space and target eigenfrequencies at different rotational speeds
- Self-programmed numerical solver for the highly-nonlinear model
- Self-programmed optimizer for the special application
Personal Projects
Multibody Simulation
An interactive web-based multibody dynamics simulation powered by WebAssembly, demonstrating real-time physics simulation in the browser.
- High-performance physics engine compiled from C++ to WebAssembly
- Real-time 3D visualization using React Three Fiber
- Interactive simulation parameters and controls
- Efficient computation using Web Workers for smooth performance
LLM Fine-tune with Llama 3.2 3B
Trump simulator 3000
- Fine-tuned Llama 3.2 3B model to simulate conversational patterns
- Created custom training dataset for personality-specific responses
- Achieved significant improvement over base model in style matching
Deep RL Arena Shooter
A custom 1v1 3D arena shooter environment built to train autonomous AI agents using Deep Reinforcement Learning. The agents learned to navigate, track opponents, and engage in intense combat entirely from scratch through trial and error.
- 4-axis continuous control space for fluid movement, camera rotation, and trigger pulls
- Custom 18-float observation array with relative spatial math, raycast sensors, and proximity radar
- Multi-layered reward function promoting tactical positioning and aggressive combat behavior
- Overcame frame-perfect synchronization challenges between Godot and Python environments
- Successfully converged after 5 million timesteps with highly reactive AI behavior