VisAutocomplete suggests next chart steps like text autocomplete
New system beats LLM vibecoding and Excel for complex chart authoring.
When domain experts create charts, the bottleneck is often knowing the optimal next step in the vast visualization design space. To address this, researchers Hyeon Jeon, Sungbok Shin, and Niklas Elmqvist present VisAutocomplete, a system inspired by text autocompletion that reconceptualizes visualization authoring as a sequential process. It recommends concrete next steps at each stage based on common practices, allowing users to intervene or delegate multiple steps and select from design recommendations. To enable responsive interaction, the team distilled the translation logic of a large language model into a single function that receives the current chart state and a recommended transition, returning the updated chart specification.
In evaluations against LLM vibecoding, Microsoft Excel, and TaskVis (an automated chart recommendation engine), VisAutocomplete outperformed all baselines in the articulacy of complex chart authoring while remaining on par with LLM in approachability. Accepted at IEEE VIS 2026 and TVCG, this system could significantly lower the barrier for domain experts to create sophisticated visualizations without deep technical expertise, making data storytelling more accessible and efficient.
- Inspired by text autocomplete, VisAutocomplete recommends concrete next steps at each stage of visualization authoring.
- Uses an LLM distilled into a single function to translate chart state and transition into updated chart specifications.
- Outperformed LLM vibecoding, Excel, and TaskVis in articulacy for complex charts while matching LLM approachability.
Why It Matters
Makes data visualization design accessible and efficient for domain experts without coding expertise.