Overview
The CSV Visualization feature enables users to upload a CSV file and generate visual insights and charts through natural language prompts. This article outlines the supported capabilities, current limitations, and known issues to help users and support teams understand expected behavior and constraints.
Supported Capabilities
The current implementation supports the following:
- Upload and analyze a single CSV file
- Generate visualizations from natural language queries
- Basic data filtering using natural language prompts
- One-turn conversational context with limited history awareness
- Visualization of categorical dimensions with up to 10 unique categories
- Support for CSV files up to 10 MB in size
File Support
- Only single CSV file uploads are supported
- Excel files (.xls/.xlsx) are not supported
- Maximum supported file size is 10 MB
Conversation & Query Handling
- Supports only one-turn conversational history
- The system does not intelligently disambiguate unclear requests before responding
- Follow-up prompts may require explicit instructions to maintain chart consistency
Data Constraints
- No CSV data validation is currently performed
- Missing data fields are ignored during analysis
- Duplicate rows are counted multiple times
- Empty lines in the CSV are ignored
- CSV files without headers may produce unpredictable results
- Some non-Unicode characters may not render correctly in generated graphs
Visualization Constraints
- Maximum supported categorical values per column is limited to 10
- Pivot table–style computations and advanced insights are not validated or officially supported
- Data filtering is supported via natural language queries
- Data sorting scenarios are not officially tested or approved
Date Format Support
- Officially supported date format: MM/DD/YYYY
- Other date formats may work but are not validated
Known Issues
The following issues have been observed during testing (tested using Llama 70B, 4-bit quantized model):
Graph & Visualization Issues
Low-Resolution PNG Downloads
Downloaded PNG images may occasionally have a resolution below 720p.
Legend Overlapping the Graph
In some cases, the legend may overlap chart elements.
Workaround: Submit a follow-up request asking to reposition the legend.
Missing X-Axis Label
Rarely, the X-axis label may not appear in the generated graph.
Double-Line Axis Appearance
The graph grid may occasionally appear too close to the axis line, creating a visual double-line effect.
This issue is cosmetic only and does not impact chart accuracy.
Incorrect Percentage Display
Percentage-based charts may sometimes display raw numeric values instead of percentages.
Workaround: Submit a follow-up request to correct the display.
Incorrect Time Granularity
A yearly chart may occasionally be generated instead of a requested monthly chart.
Workaround: Submit a follow-up request specifying the desired granularity.
Graph Type Changes During Follow-Up Requests
Follow-up prompts may unexpectedly change the chart type (for example, from Pie Chart to Bar Chart).
Workaround: Explicitly instruct the bot to retain or restore the desired chart type.
UI & Workflow Issues
File Panel Closing Unexpectedly
Deselecting a file to choose another one may unexpectedly close the file panel.
Thread Rename Failure
Renaming the current thread to a custom name may occasionally fail to apply successfully.
Timeout Handling Inconsistency
If a user request exceeds the 90-second timeout limit:
- The end user may see an error message
- The request may still incorrectly appear as “Resolved” in the Requests tab
Classification Issue with Small Talk Enabled
If the bot/application has Small Talk enabled, some valid CSV visualization requests may be incorrectly classified and result in fallback responses.
Recommended Workaround
Temporarily disable the Small Talk feature in the bot/application configuration.
Best Practices
To improve visualization reliability and accuracy:
- Use CSV files with clear column headers
- Keep categorical values under 10 unique entries per dimension
- Use the supported date format: MM/DD/YYYY
- Avoid unsupported or special character encodings
- Use explicit follow-up instructions when refining charts