Instant Charts: The Magic of Translating Data to Charts

Discover how at Inpla we transform natural language questions into automatic graphs with AI and Vega-Lite, accelerating decision-making without depending on experts.

When the table shouts at

you

Interpreting millions of rows without a chart is a headache.

Pikachu meme with surprised expression (mouth and eyes wide open) on a blurred green background, with a screenshot of a spreadsheet (columns: Product, Quantity, Sale Amount, Region, Seller, Date, Cost) superimposed on top.

A text-only report leaves you more confused than before.

Natural language + Graphics = Frictionless

decisions

According to Gartner, natural language for BI is catching on because it democratizes access to business intelligence.

Now anyone can ask "What was my budget this month?" and receive a chart, without using Power BI or Excel.

The secret combo: Vega-Lite and LLMs

We outline the columns: we identify categories, numbers and dates.
LLM recommendation: we ask the model to suggest chart types (bar, line, pie, heat maps, etc.) compatible with Vega-Lite.
Choice by trust: we select the option with the highest confidence index.
Final Specification: We generate the Vega-Lite JSON that produces a clear and attractive graphic.

From theory to action

In theory, we were inspired by the papers Chat2VIS and Chat2Data, which demonstrate how graphs can be created from data.

Based on this, Diego and José designed a flow that, based on a JSON and instructions, allows the AI to recommend and create graphs.

For all the details, listen to the full

episode

In our YouTube podcast, we tell how we started from those papers, chose Vega-Lite and designed the flow with the LLM to recommend the perfect graphic. During the episode we do a live demo of how, from a business query such as "What are the best-selling products?", a graph is automatically generated.

Watch the podcast here: https://youtu.be/dvn2LzJP3IA