Large language models (LLMs) have a hard time working with CSVs because the data lives far away from the labels.
Here's an example of what that means: In this example, let's imagine you have a massive spreadsheet of data -- like API request response times. You might want to ask Storytell a question about the data, like "What was TechAnalytics' request time in November 2021?". But the title label of that column and description label of that row are far away from the number; there are many numbers in-between. When an Storytell (or any LLM) tries to parse the data, it doesn't know to attach the column "November 2021" and row "Tech Analytics" labels to answer.
Here's how to fix this: Convert your .CSV into an .XML file
While you can't upload .csv files into SmartChat™, you can upload XML files. XML is a structured markup language that wraps the labels around the data. We recommend using OnlineCSVTools.com. Here's how it works:
Here's what your .CSV file looks like once you've converted it to XML: You can see the first <November 2021> label now wraps the data: