Yu Wang

About Me

Hi there, I'm Yu Wang, just got my Ph.D. degree from Visualization and Graphics Group, Utrecht University, under the supervision of Prof. Alexandru Telea. My research interests include high-dimensional data visualization, adversarial training (GAN), and machine learning classifier visualization.

(click to see my PhD tree)

Alongside research, I now work as a software engineer and have been building practical AI tools around retrieval, interactive interfaces, and LLM-assisted workflows. More recently, I have been leaning toward agentic-AI-related product and engineering work.

Fun fact: I am an FGA-certified gemologist certificate with expertise in gemstone identification.

Research

Visual analytics for high-dimensional data and machine learning

I build methods and interfaces that make embeddings, inverse projections, and classifier behavior easier to inspect instead of treating them as black boxes.

Most of my work sits between dimensionality reduction, explainable AI, interactive systems, and practical AI tooling. The projects below mix publication-linked research software with side projects that reflect how I build user-facing systems.

Inverse projections Decision maps Interactive ML systems Research software Applied LLM tools

Selected Research and AI Projects

Repos that connect papers, prototypes, demos, and a few practical AI side projects.

Browser demo

Browser Decision Map

Browser-native MNIST demo that loads TensorFlow.js models client-side, renders decision regions over the embedding, and lets users probe inverse projections directly in the map.

TensorFlow.js D3 WebGPU MNIST

Try demo View repo

Research software

LCIP (AKA how to make dogs smile)

Controllable inverse projection system for high-dimensional image data. LCIP adds a user-controlled latent direction so exploration is not limited to a single fixed reconstruction surface.

PyTorch PySide6 StyleGAN2 Visual analytics

Read paper View repo

Python package

InverseProjections

Reusable toolkit for inverse projection workflows, collecting NNinv, iLAMP, RBF interpolation, multilateration-based inversion, and gradient-map analysis behind a consistent API.

pip install inverse-projections

Python PyTorch scikit-learn Gradient maps

View repo

Decision-map research

Generalized FastDBM

Framework for computing dense explanation maps for classifiers over UMAP and t-SNE embeddings using inverse projection plus adaptive refinement, so expensive maps can be approximated far faster than exhaustive grid evaluation.

PyTorch TensorFlow Adaptive refinement Decision boundary maps

shell_ai

Minimal local shell agent that turns natural-language requests into a single command, prints a short reasoning trace, and asks for confirmation before running anything on the host machine.

uv run ai "find the 20 largest files in this directory"

Python pydantic-ai Gemini CLI

View repo

Applied AI side project

RAG Thesis System

A chat interface for my PhD thesis, built after friends told me the PDF was too long and difficult to read. It combines retrieval, reranking, citation handling, memory, and a browser UI into a compact LLM application that also reflects my growing interest in agentic AI tooling.

FastAPI ChromaDB Hybrid retrieval LLM app

Read thesis View repo

Selected Publications

Hi, this is Yu Wang