Péter Ferenc Gyarmati
I’m doing my PhD at
I study how people can guide analytical work with AI, correct it, carry their judgment forward, and account for its information and privacy consequences. I also build tools for
Before my PhD, I built software in industry.
Questions I keep coming back to
Select a question
to see the work.
chartcoach
chartcoach collects 781 visualization guidelines and keeps each source and exception beside the advice. People and chart tools can search the same catalog.
About the project →
marimo-lens
marimo-lens records a marked notebook result, note, image, and producing cells as one request. The request stays intact while an agent works and a reviewer checks the result.
About the project →
marimo-studio
marimo-studio turns a marimo notebook into web pages with their own layouts and interactions. The pages reuse the notebook's Python data, calculations, and controls.
About the project →
VACP
VACP exposes a data-driven app's structure, inspectable data, and controls to an AI agent. After an action, the agent can inspect the resulting view.
About the project →
Do AI models see charts like people do?
Thirteen image-capable models recognized a figure's purpose and dimensionality more reliably than its visual marks. Expert annotations across 305 scientific figures provided the comparison.
About the project →
Agentic Visual Reporting
Agentic Visual Reporting composes language models with statistical tools and explicit chart-design rules. A fixed pipeline turns their outputs into an interactive report with executable provenance.
About the project →
anywidget-mcp
anywidget-mcp turns existing AnyWidgets into MCP Apps, reusing their Python logic and JavaScript interfaces inside AI conversations.
About the project →
pyobservablejs
pyobservablejs brings Observable’s reactive cells and browser libraries into Python notebooks. Python supplies the data, while JavaScript inputs and charts update together in the browser.
About the project →
Draco 2
Draco 2 checks charts against explicit design rules, completes partial specifications, and ranks alternatives. Each result shows which rules shaped its score.
About the project →