How It Works

1

Upload

Upload your CSV file to Storytell. For detailed instructions on uploading content, see our Uploading Content guide.

2

Processing

Storytell processes the first 1,000 rows of the CSV file

3

Story Tile™ Generation

Each row is converted into a Story Tile™.

4

AI Analysis

Storytell LLM analyzes the Story Tiles™.

5

Insight Extraction

Query your data using SmartChat™.

As you can see in the image above, your uploaded CSV files are easily accessible in the “Stored Assets” section. You can start querying your data immediately after upload.

Generating Story Tiles™

Story Tiles™ are clusters of related concepts from your data. For example, let’s say you have a CSV file with the following information:

NameCompositionDistance from the Sun (AU)Orbital Period (years)Diameter (km)
MercuryRocky0.390.244879
VenusRocky0.720.6212104
EarthRocky1112742
MarsRocky1.521.886779
JupiterGas Giant5.211.86139820
SaturnGas Giant9.5829.46116460
UranusIce Giant19.2284.0150724
NeptuneIce Giant30.05164.7949244
MoonRockyN/A0.0743474
TitanIcyN/A15.945149

Here’s what a Story Tile™ would look like:

“Titan” is a “Icy” celestial body located “N/A” AU from the Sun with an orbital period of “15.94” years.It has a diameter of “5149” kilometers.

This transformation enables our AI to understand relationships and context within your data, making it possible to answer complex queries.

Querying Your Data

With Storytell, you can ask questions about your CSV data and receive clear, insightful answers through SmartChat™:

As demonstrated in the image above, when you ask about Titan’s orbital period, Storytell provides a precise answer based on the processed data.

Verifying Accuracy

Storytell’s responses are based on the data you provide. You can always verify the information by checking the original CSV file:

This screenshot confirms that Storytell’s response matches the data in your original CSV file.

Technical Considerations

  • Processing limited to first 1,000 rows for speed and efficiency
  • Secure, isolated environments for data privacy
  • Scalable architecture for concurrent processing

Handling Structured vs. Semi-Structured CSV Files

The Storytell process involves classifying CSV file content as either “structured” or “semi-structured” to determine the appropriate processing strategy. This classification is crucial for handling CSV files that do not conform to traditional tabular formats, ensuring that data is processed accurately and efficiently.

Structured vs. Semi-Structured Data

Structured Data

Structured data in CSV files typically includes a clear header row followed by consistent data rows. Each column represents a specific data attribute, and each row contains data entries corresponding to these attributes. This format is straightforward to process using standard CSV parsing techniques.

Semi-Structured Data

Semi-structured data, on the other hand, may not have a consistent structure. These CSV files might lack headers, have inconsistent columns, or contain data that resembles reports rather than traditional tables. Such files require a different approach to ensure accurate data extraction and processing.

Process for Classifying CSV Content

  1. Initial Inspection: Storytell begins by automatically inspecting the CSV file. During this phase, the system determines key characteristics, such as the presence of a header row and the consistency of data rows throughout the file.
  2. Classification: Based on the results of the initial inspection, Storytell classifies the CSV content into one of two categories:
    • Structured: If the CSV file contains a clear header row and all data rows are consistent, it is classified as structured data.
    • Semi-Structured: If the file is missing a header or exhibits inconsistent columns, it is classified as semi-structured data.
  3. Prompt Selection: Based on the classification, Storytell selects the appropriate processing prompts tailored to the content type:
    • For structured data, Storytell utilizes standard CSV processing prompts to ensure efficient data handling.
    • For semi-structured data, the system employs specialized prompts designed to manage variability and generate data “chunks.” Key considerations include:
      • Ensuring that the language model does not drop any data points by refining the prompts.
      • Including a “source” identifier in each chunk to enhance data searchability and retrieval.

Multi-Tab XLS Files

Large language models (LLMs) often struggle with interpreting data from multi-tab XLS files due to their reliance on proximity to make sense of information, which can lead to confusion, especially with complex datasets. Storytell addresses this challenge by converting data into a more coherent format.

Storytell currently supports up to 20 tabs for multi-tab XLS files. Ensure that your file does not exceed this limit.

Storytell’s Process for Multi-Tab XLS Files

1. Uploading the File

When you upload a multi-tab Excel file into Storytell, the system processes each tab’s data individually.

2. Breaking Down the Data

Storytell breaks the uploaded file into discrete pieces of information, analyzing each row and column to identify key concepts and relationships.

3. Creating Story Tiles™

After breaking down the data, Storytell converts the rows into Story Tiles™, coherent sentences that represent the information for easy understanding by LLMs.

This transformation allows LLMs to provide accurate responses based on the data you’ve uploaded.

4. Interacting with Your Data

After uploading your multi-tab Excel file, create a Collection to organize your data. You can then interact with this Collection and ask specific questions, with Storytell providing clear answers using generated Story Tiles™.

For example, you can pose a question that needs information from two separate tabs:

Sample: “Tell me how the orbital period of Titan compares to the orbital period of Planet 164.”

Storytell will pull the relevant information from your multi-tab XLS data.

Learn more about the process by watching DROdio’s video demonstration that highlights its functionality.

Future Enhancements

  • Support for larger datasets (beyond 1,000 rows)
  • Advanced data type detection and custom Story Tile™ generation