Visistant++: Enhancing Conversational NL-to-Visualization with Voice Input and Multi-File Join Capabilities
Masetty Jyothi Ram Swaroop, Kukkapalli Venkatanaga Sai, Pamidi Santosh Kumar, Dr. M. Sridhar · Iconic Research and Engineering Journals · 2026
Natural Language to Visualization (NL2VIS) systems make data analysis accessible to everyday users without coding skills. Visistant, built on Google’s Gemini large language models and Streamlit, demonstrated strong performance in generating interactive Plotly charts from typed natural language queries. However, two practical barriers limited its real-world adoption: users had no option to speak their queries instead of typing them, and analysts could not combine columns from multiple uploaded CSV files in a single query. This paper presents Visistant++, which addresses both gaps directly. We add a Voice Input Module that captures spoken queries through the browser-native Web Speech API and presents the transcript for confirmation before submission. We also introduce a Multi-File Join Engine that lets users configure SQL-style joins (inner, left, right, outer) across two uploaded datasets on a shared key, after which the merged data is queried as a unified dataframe. Both features integrate cleanly into the existing Gemini pipeline without altering the core prompt or code-generation architecture. We describe the design, implementation, and observable behaviour of each enhancement and discuss their contribution to making NL2VIS tools genuinely accessible to a wider range of users.