About the Gen AI Tool
The Savant Gen AI Tool, currently in beta, offers generative artificial intelligence capabilities within your Savant analysis. This tool leverages the power of Large Language Models (LLMs) to enhance your data analysis process. Supported LLMs include OpenAI's ChatGPT, Google Gemini, and more.
Configuring an LLM Service
Configure Your OpenAI Account
Savant supports the integration of LLM environments such as OpenAI and Azure. A default Savant AI account is available for minimal testing, but for more comprehensive usage, you can configure your own LLM service provider.
Steps to configure OpenAI:
Navigate to the Systems page.
Click New System.
Search for OpenAI and select it.
Choose the environment: OpenAI or Azure.
Provide your API token from the selected environment and click Authenticate.
Rename and describe your OpenAI provider.
Confirm the setup.
For access to Google Gemini, please contact us to join our private beta.
Configuration Options
Prompt Configuration
LLM Service Selection: Choose which LLM service you would like to use (e.g., Savant Demo, OpenAI, etc.).
Define Your Prompt: Write a well-structured prompt that specifies the analysis you'd like the LLM to perform. You can include detailed instructions for generating relevant insights from your data.
Example Prompt: "What country is this city found in? Return just the name of the country with no other words. If you cannot guess the country, return nothing."
Field to Transform: Select the field in your dataset that you want to transform based on the prompt.
Settings Configuration
Max New Rows Processed per Run: Set the maximum number of rows processed each time the configuration is run. This is designed to conserve LLM credits. The default value is 1000.
Processing Mode:
Batch Rows for Large Data: Use this mode for large datasets, where rows are processed in batches.
Stream Rows for Small Data: Use this mode for smaller datasets, typically less than 100 rows, for faster run times.
Additional Notes
Output
Once you apply the configuration, the following changes will occur:
In development mode, Savant will calculate 5 new records each time you click Apply.
If you click Apply multiple times, Savant will calculate 5 additional records until it reaches 1000 records for that prompt and data.
Running the entire bot calculates all data, with 1000 records visible in the development mode preview. This is designed to help conserve your LLM service credits.
The output from the LLM service will be added as a new AI Answer field in your dataset.
Referencing Columns in Streaming Mode
In advanced configuration, you can use Streaming Mode for smaller datasets (typically less than 100 rows) with faster run times.
This mode also allows dynamic injection of fields into your prompt. To reference a field, simply include ${field name} in your prompt.
Example: For a record where City = "Paris" and Country = "Japan", the prompt "Is ${City} in ${Country}?" would query the LLM with "Is Paris in Japan?"
Importance of Adding Specific Instructions
To maximize the accuracy of the AI-generated insights, it’s crucial to provide clear and specific instructions. Avoid ambiguity by being explicit in your prompts, ensuring the AI returns precise and relevant results based on your analysis needs.