Luminal

Every action started with a sentence. You described the change you wanted and Luminal mapped it to an operation on the file. No formula syntax, no menu hunting, and no need to remember which function handles text splitting.

The tool handled the boring repairs: duplicate rows, inconsistent capitalization, mixed date formats, trailing spaces, and categories spelled three different ways. You could also ask it to reorganize columns into a different order. Cleaning was the job most users came for, and it was the part the product was built around.

Luminal didn't just pattern-match text. It generated Python code and ran it against your data, which is why it could handle calculations and reshaping that a simple find-and-replace tool can't. Directory listings from the company's launch period say the first version relied on GPT-3 for that code generation. The tradeoff is familiar: generated code can be wrong in ways that look plausible.

Instead of a pivot table, you asked a question. The pattern is the same one behind most natural language data analysis tools: describe the outcome, let the model pick the method. Which region had the highest return rate, what the average order value was, how many customers ordered twice. Luminal answered in text and, when asked, produced a table to back up the number.

A written request could produce a bar chart, a line chart, or a map from geographic columns. That covered the reporting layer for people who don't want to leave the browser for a separate visualization tool.

Prompts worked in several languages, including Spanish, French, Japanese, Hindi, and Chinese, which mattered for teams working across markets. The company said files were encrypted in transit and at rest, hosted on major cloud infrastructure, with no long-term retention of customer data. Those are vendor claims, and they're worth confirming before you upload anything sensitive.