Software Engineer · AI Engineer · Explainable ML · Visual Analytics · Agentic AI

Dual Ph.D. background in Computer Science and Geology. I build research software that makes complex models, high-dimensional data, and scientific workflows easier to inspect, reproduce, and use.

Website · Google Scholar · Repositories · GitHub

I am based in Utrecht, Netherlands, with a Ph.D. in Computer Science from Utrecht University and a Ph.D. in Geology from China University of Geosciences, Beijing. My work connects machine learning, visual analytics, geoscience, and research-oriented software development: implementing methods from papers, building usable prototypes, and turning research workflows into software that other people can run and inspect.

I am especially interested in research software engineering roles where software quality, reproducibility, and collaboration across scientific domains matter. Recently, I have also been focusing on agentic AI, including retrieval-augmented generation, tool-using agents, and AI-assisted workflows. My broader research interests include explainable machine learning, human-in-the-loop data generation, inverse projection, and decision maps.

• Ph.D. in Computer Science, Utrecht University: enhanced decision maps for exploring classification models.
• Ph.D. in Geology, China University of Geosciences, Beijing: machine learning and visualization for mineral-genesis classification.
• Publications in venues and journals including Computers & Graphics, American Mineralogist, IVAPP/VISIGRAPP, Algorithms, SN Computer Science, and JGR: Solid Earth.
• Best Student Paper Award at IVAPP/VISIGRAPP 2024.
• Open-source contribution to SHAP.

For a quick review of my research software work:

• Best RSE example: LCIP shows tested Python research software, GUI tooling, CUDA/PyTorch workflows, and reproducibility scripts.
• Best Earth-science example: SDBM for Pyrite is a reproducible geoscience ML workflow linked to an American Mineralogist paper.
• Best reusable-package example: InverseProjections exposes inverse-projection methods through a scikit-learn-style API.

Loss-Controlled Inverse Projection of High-Dimensional Data. This is my strongest current research-software project: a paper-linked Python implementation with a Qt GUI, command-line entry points, tests, CUDA/PyTorch-based demos, reproducibility scripts, and documented workflows for inverse projection experiments.

In plain terms, LCIP supports human-in-the-loop, visually guided generation of high-dimensional data from a 2D embedding. It connects visualization, generative modelling, and interactive model steering.

Signals: research software engineering, scientific visualization, human-in-the-loop generative ML, interactive tooling, reproducibility, GPU-enabled workflows.

Reproducibility workflow for interpreting mineral-genesis classification with supervised decision maps on pyrite trace-element data. The project combines geoscience data, classifier evaluation, SSNP-based projection, inverse feature mapping, and notebook-generated manuscript figures.

Signals: Earth-science-facing ML, geochemistry, explainable classification, visual analytics, reproducible computational research.

Implementation accompanying research on fast and accurate decision maps for explaining classification models.

Signals: explainable AI, model inspection, visual analytics, research-method implementation.

A Python package implementing inverse projection techniques such as NNinv, iLAMP, RBF inverse mapping, and MDS multilateration with a scikit-learn-style API.

Signals: reusable research code, dimensionality reduction, interpretable high-dimensional data analysis.

A lightweight local shell agent that turns natural-language requests into executable shell commands, explains the proposed command, and asks for confirmation before running it.

Signals: LLM tooling, CLI user experience, safe tool use, practical agent design.

Exploration of retrieval-augmented generation workflows and knowledge-aware LLM application design.

Signals: retrieval, context construction, LLM application engineering.

A browser-based interactive decision-map demo using TensorFlow.js and D3 to inspect MNIST projections, inverse projections, decision regions, and observation windows directly in the browser.

Signals: interactive visualization, browser ML, scientific demos, user-facing research prototypes.

Languages:        Python · JavaScript · SQL
ML/data:          PyTorch · TensorFlow · scikit-learn · pandas · NumPy · XGBoost
Visualization:    Matplotlib · seaborn · D3.js · PySide/PyQt · pyqtgraph · vispy
RSE strengths:    Reproducibility · Scientific Workflows · Documentation · Interactive Tools
AI systems:       Agentic AI · RAG · LLM Applications · Tool Use · Workflow Automation
Engineering:      FastAPI · SQLAlchemy · Pydantic · pytest · PostgreSQL · TensorFlow.js

I am interested in research software engineering, AI engineering, and data science roles, especially in teams that build software for scientific research, environmental and Earth-science applications, visual analytics, explainable AI, agentic AI, or knowledge-intensive workflows.

Website: http://yuwang-vis.github.io/

GitHub: https://github.com/wuyuyu1024