Chat your way to charts instead of wrestling drag-and-drop

What it does

VizGPT is a chat interface that generates data visualizations from tabular datasets using GPT. You upload a CSV, describe what you want in plain English, and get a Vega-Lite chart back. The hook is iterative refinement: you can ask follow-up questions, tweak colors, adjust axes, or pivot the analysis without retyping your entire request.

The interesting bit

The project treats visualization as a conversational loop rather than a one-shot generation task. The README explicitly pitches this against tools like Tableau where “making a heatmap requires bin transformations to both axes and then dragging the measure to color” — a friction point for non-specialists. There’s also a RAG-augmented version that retrieves Vega-Lite and SQL documentation to ground the model’s output.

Key highlights

• Outputs Vega-Lite specs, which are portable and declarative
• Maintains chat context for step-by-step visual exploration
• Auto-infers dataset metadata (columns, types) with manual override
• RAG variant available at the hosted playground for better spec accuracy
• Explicitly scoped to visualization, not data transformation or cleaning

Caveats

• Saving visualizations and chat history is on the roadmap but not implemented
• Requires Azure OpenAI credentials; no mention of other providers
• The hosted playground’s long-term availability depends on the maintainer’s infrastructure

Verdict

Worth a look if you frequently sketch exploratory charts and find GUI tools tedious, or if you’re building analytics features for users who won’t learn a visualization grammar. Skip it if you need heavy data preprocessing — the README nudges you toward Kanaries/RATH for that.