Zhengxuan Yuan
Zhengxuan Yuan / Applied AI Engineer Open to work
Engineering × Intelligence

Reliable AI systems.

For industrial data, simulation,
and agent workflows.

I turn research prototypes into dependable engineering workflows for real technical data.

Explore projects About me ↗
Intelligence, applied01 / 03
CAE / GRAPH REPRESENTATION
RESPONSE− / +
PreviouslyApplus+ IDIADA
DisciplineApplied AI
Based inMunich, DE
Simulation · Signals · Agent systems
01 / Selected work

Built around real problems.

A selection / 3 projects
01 — LLM workflows / Professional R&D

Case-to-Skill
Agent

Knowledge that becomes capability.

Case-to-Skill Agent turns solved engineering cases into reusable, evaluated procedures. Retrieval is only the beginning.

70% → 80%Overall target score on a public engineering benchmark.

GNN Surrogate Modeling

GNN surrogate modeling for crash simulation. Prediction grounded in engineering evaluation.

↗
GNN Surrogate Modeling / Read project notes

Research and engineering workflow for crash simulation prediction, evaluation, and validation. A selected custom-loss experiment reduced HIC error by 27%; this result applies to that experiment, not every model or scenario.

Read full case study

arTIco — CAE Validation

arTIco learns expert assessments from crash-simulation signals to support digital-twin validation.

arTIco / Read project notes

Engineering validation with 5-fold cross-validation and 1,000 bootstrap resamples. The visualization illustrates the learning and evaluation workflow; the displayed chart and matrix are schematic.

Read project overview
02 / Project archive

Complete engineering catalog.

17 projects / 7 disciplines

Project selector

Find your field of interest

01 / ALL17 matched
All projects
Matched projects
01-05 / 17
Professional work6 projects
  1. P-001Agent Systems / Case-Based ReasoningCase-to-Skill AgentInstead of returning isolated passages, the system retrieves validated procedures that can be adapted, executed, evaluated, and safely retained.
  2. P-002Graph Neural Networks / Scientific MLGNN Surrogate Modeling for Crash SimulationAn applied GNN surrogate-model research and engineering workflow for crash simulation prediction, evaluation, and validation.
  3. P-003Computer VisionEV Charger DetectionAn end-to-end computer vision pipeline for collecting, filtering, labeling, and classifying EV charger imagery.
  4. P-004Time-Series Machine Learning / Model ValidationarTIcoA research ML framework for learning expert assessments from multichannel crash-simulation signals to support digital-twin validation.
  5. P-005Binary ClassificationBroken Signal DetectionA modular ML workflow for classifying time-series measurement channels as OK or NOK and supporting the review of abnormal signal patterns.
  6. P-006MLOpsCrash Test SignalAn AWS Lambda inference service for validating crash-test sensor channels with real-time signal processing.
Personal tools7 projects
  1. P-007Conversational BI / Agent SystemsAI-BIA full-stack retail BI system that combines a live dashboard, LangGraph tool orchestration, structured analytical artifacts, and a four-tier offline-capable data layer.
  2. P-008macOS Utilityds-monA macOS menu bar utility for monitoring DeepSeek API balance with secure local configuration.
  3. P-009Finance AITradingAgents UIRun multi-agent stock or crypto research, follow the evidence live, and reopen or export the finished report.
  4. P-010macOS AI UtilityMac ASR + TranslateA native macOS command-line workflow and reusable Agent Skill for transcribing local audio or video, translating the result, and producing timestamped subtitle artifacts.
  5. P-011macOS AI Plugindsh-mac-visionA DeepSeek Harness plugin for inspecting local images, clipboard content, screens, and application windows through on-device Apple Vision.
  6. P-012Fintech Workspacedsh-trading212A read-only Trading 212 portfolio workspace for dsh with live performance tracking, ECharts price history with buy/sell fills, and conversational tools.
  7. P-013Career IntelligenceJob Search DEAn agent skill that collects job listings, checks source availability, compares requirements with candidate evidence, and exports a workbench for reviewing roles and tracking applications.
Academic research4 projects
  1. P-014Scientific Machine LearningMaterial ModelA FEM–ML workflow for identifying YLD2000-2d material parameters from experimental full-field strain responses.
  2. P-015Computational EngineeringSpringback CompensationAn iterative FE and geometry-processing workflow that converts springback deviations into a corrected tool surface for the next forming simulation.
  3. P-016Computer VisionCellVisionA blood-cell analysis platform spanning segmentation, labeling, active learning, correction, and reporting.
  4. P-017Computer VisionTour Into the PictureAn interactive single-image 3D reconstruction using vanishing-point geometry and a five-plane room model.
03 / The engineer

Engineering first.
Intelligence, applied.

My work sits where machine learning meets physical systems. I build the evaluation, tooling, and workflows that help research hold up against real engineering data.

  1. 2023 — 2026Professional experience

    Applus+ IDIADA

    Applied AI Engineer

    Munich, Germany
  2. 2020 — 2023Master’s degree

    Technical University of Munich

    M.Sc. Mechatronics and Robotics

    Munich, Germany
  3. 2014 — 2020Bachelor’s degree

    Leibniz University Hannover

    B.Sc. Mechanical Engineering

    Hanover, Germany
Selected work / 01

Case-to-Skill Agent

A case-based LLM workflow that converts solved engineering cases into reusable, evaluated skills.

My contribution
Co-engineered selected workflow components: retrieval and context design, structured outputs, evaluation loops, fallback paths, caching, and cost-aware orchestration.

The engineering decision
Retrieve validated procedures rather than isolated passages. Evaluate generated results against objective criteria before retaining a skill; use retry or fallback when needed.

Reported result
70% → 80% overall target score on a public engineering benchmark.

The homepage animation illustrates the workflow. It does not expose client data, internal prompts, or production execution.