AI Agent

From QPR ProcessAnalyzer Wiki
Revision as of 16:01, 2 March 2025 by Ollvihe (talk | contribs) (→‎Take AI Agent into Use)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

The AI Agent is an innovative dashboard component designed to display textual responses generated by language models based on user-configured prompts. This component seamlessly integrates with both case-centric and object-centric Snowflake models, to provide comprehensive insights.
Key features of the AI Agent include:

  • Versatile Model Support: The AI Agent is compatible with language models from both OpenAI and Snowflake Cortex, offering flexibility and a wide range of capabilities to meet diverse analytical needs.
  • Dynamic Response Generation: By configuring specific prompts, users can tailor the AI Agent to generate relevant and context-specific textual responses, enhancing decision-making processes.
  • Integration with Snowflake: The AI Agent supports Snowflake's case-centric and object-centric models, allowing users to harness the full potential of their data within the Snowflake ecosystem.

This component is ideal for users seeking to leverage advanced language models to extract meaningful insights and drive data-driven strategies directly from their dashboards.

Take AI Agent into Use

To use the AI Agent, Snowflake needs to be configured, so that the Snowflake LLM's are available for the AI Agent. Optionally, the OpenAI language models can be used if the OpenAI connection is configured. To use OpenAI default model, leave the LLM name field empty. See Snowflake Cortex or OpenAI related settings respectively.

You can also fine-tune the large language model, so that it better suits to your use. For example, the large language model can be fine-tuned with background information regarding the analyzed process and the broader business environment. For more information about fine-tuning large language models, please contact QPR or check out the Snowflake Cortex Fine-tuning or OpenAI Fine-tuning technical instructions.

In Snowflake, it's possible to restrict which language models can be used. By default, all language models that are available in the same region, can be used. More information how to restict the available language models: https://docs.snowflake.com/en/user-guide/snowflake-cortex/llm-functions#control-model-access.

To use Snowflake language models that are available in other Snowflake regions, the Cross-region Inference feature can be used. More information: https://docs.snowflake.com/en/user-guide/snowflake-cortex/cross-region-inference.

General Settings

  • System Prompt (optional): Enter a system prompt for the language model if desired. This can help guide the model's responses but is not required. The system prompt can be used give general instructions how the LLM should behave when responding to the user.
  • User Prompt: Provide a user prompt for the language model. This field must contain valid input. Usually, the user prompt is used give the specific task or question the language model should respond to.
  • Input Data (optional): Optional. Configure an array of zero-to-many chart configurations to be used as source data for the language model. The AI Agent can function without input charts. You can copy chart settings from another chart and and use them here as follows: Open the Chart Settings (editable) of the other chart, select all JSON and copy it to a clipboard. Come back to the AI Agent Input Charts, paste the settings to the textbox, and click Apply. Note that if the input chart's data exceeds 20000 characters, the data is cut at the end of the line where the limit is reached.
  • LLM Name: Select a language model from the predefined list of Snowflake Cortex models or enter a custom model name (for example your own fine-tuned model). If the OpenAI connection is available and no LLM name is specified, the OpenAI language model will be used by default.
  • LLM Parameters (optional): Define optional parameters for the language model using a JSON object. Refer to the Snowflake and OpenAI documentation for supported parameters. Example: { "temperature": 0 }
  • Title (optional): Custom title for the AI Agent can be defined. Note that the AI Agent does not have an automatically created title.
  • Description (optional): Provide a description for the AI Agent. Note that this description is not used as a prompt for the language model.

Layout Settings

Following layout settings are available:

  • Background color: Background color of the AI Agent area. The color can also contain partial transparency (alpha value) or even be fully transparent. When the A Agent background has transparency, the color of the dashboard background is visible behind the AI Agent.
  • Border color: Border color of the AI Agent. Like the background, border color can also contain transparency.
  • Border width: Border width of the AI Agent in pixels. When the width is zero, border is not visible.
  • Border corner roundness: Border corner roundness in pixels. Zero means sharp corners.

Filtering Settings

  • Follow Dashboard Filters: When checked, the AI Agent is filtered by the filters in the dashboard. When unchecked, the AI Agent is not affected by filters in the dashboard, and thus the AI Agent feedback is based on data in the entire model.
  • Chart Filter: Apply filters to each of the input charts if they are set to Follow Dashboard Filters.
  • Find Root Causes: Rules to select cases used by the root causes analyses to find possible root causes for the phenomena pointed by those rules.
  • Require Root Causes Criteria: When enabled, the chart will show a user-friendly message if no criteria is selected for the Find Root Causes selection. This setting can be enabled for the AI Agent performing the root causes analysis, i.e., the root causes selection is mandatory for the AI Agent to show relevant results.

Advanced Settings

  • Settings Available in Preview Mode: If there is no need for a user to change the AI Agent settings when viewing the dashboard, this option can hide the settings in the preview mode. This helps to make the dashboard cleaner looking as there is no settings button in the AI Agent. Then changing the AI Agent settings can only be done in the edit mode.
  • On-Screen Settings: See On-screen settings.
  • Linked Settings: See Linked settings.
  • Linked Settings Disabled: Can be used to temporarily disable Linked Settings.
  • Model: Select a model for the input charts. This can be overridden in each chart's configuration.
  • Object-Centric Perspective Settings: Configure these settings if an object-centric model is selected. These settings can also be overridden in each chart's configuration.
  • Event Type Mapping: Select the event attribute used as the event type for the input charts. The event type selection affects the process flow and for example variations and flows. When the event type mapping is defined, the input charts work similarly as if the same event type mapping had been selected in the model level. When <model default> is selected, the event type mapping defined for the model is used. Note that filters don't contain the event type mapping information, so they use the mapping that has been defined for the input charts, and thus charts with different event type mapping don't work together with filtering (usually all cases appear to be filtered out due to the mismatch).
  • Chart Settings (Editable): All settings in the AI Agent are stored in a single entity that are shown here. The settings can also be edited directly. There are validations in place, so invalid settings are not accepted.

Data Sent to LLM Providers

AI Agent is based on the large language models (LLM) offered by Snowflake and OpenAI through their APIs. When using Snowflake LLMs, the data is already in Snowflake. See Data Sent to OpenAI for information what is sent to OpenAI when its LLMs are used.

Retrieved from "https://wiki.onqpr.com/pa/index.php?title=AI_Agent&oldid=25938"