Admin > Generative AI
Intended audience: DEVELOPERS ADMINISTRATORS
AO Platform: 4.3
Overview
The Generative AI section provides pages for master data configurations used by the various Generative AI functionalities in the AO Platform.
Most of the pages in this section offer the following general actions:
Search/Filter - use a search term to find an entry from existing configurations.
Delete - batch deletion of any/all selected items.
Add New - create a new entry on the page.
Refresh - refreshes the page.
Import from Excel - allows the provision to import the configuration of one or more entries of the selected configuration page.
Export as CSV - exports the content of the page to a CSV file.
Settings - allows configuration of which columns can be seen on the page.
Go to (or select by page number) - allows moving to another page in case of too many entries for one page.
Additionally, these are the actions on individual entries:
View - views the properties of an entry in read-only mode.
Edit - allows the properties of an entry to be edited.
Delete - deletes an entry.
Add to Transport - adds the configuration record to Transport.
In addition to the master data configured and stored in the configuration tables in the AO Platform, it’s important to understand both how the AO Platform handles data in the integrated LLM architecture and workflows, as well as the costs associated with using external LLM platforms, such as OpenAI. See…
Generative AI > Prompt Templates
The Prompt Templates page includes configured Prompt Templates for the different sub-systems of the AO Platform when enabled to use a Large Language Model to generate a Natural Language Generated response, including Text2SQL, Quick Insights, Data Summarizations for Apps on Results and Dashboards pages, etc…
Generative AI > LLM Configurations
The LLM Configurations page includes the configuration records for the Large Language Models that have been integrated with the AO Platform. Some LLMs are commercial LLMs and therefore may be subscription-based, whereas other LLMs are based on Open Source. Please understand the license and use terms before integrating and enabling with the LLMs.
Generative AI > Embedding Models
The Embedding Models page includes configuration records for the Embedding Models used by the different LLMs Models available and configured on the LLM Configurations page.
Generative AI > LLM Tasks
The LLM Tasks page includes configuration records for the various LLM Tasks used by the AO Platform to execute the functionality relating to the use of LLMs in the various sub-systems of the AO Platform, including Test2SQL, Quick Insights, Data Summarizations, etc…
Generative AI > Ontology Word Prompts
The Ontology Word Prompts page includes functionality that enables the establishment of criteria or conditions tailored to a particular Ontology, centered around a designated keyword, word, or phrase. The "LLM Prompt Instruction" furnishes comprehensive guidelines applicable to any prompt within the Ontology Word Prompt framework.
Example: RegionInstructionInteractiveMsoQuery:
When the region is specified with the state, it refers to the address.city property from MSO.
When the region is specified with the country, it refers to the address.state property from MSO.
When the region is specified with the city, it refers to the address.postalCode property from MSO.
Generative AI > Ontology MSO Prompts
The Ontology MSO Prompts page includes the configuration of LLM Prompt Instructions and provides detailed guidelines for each unique combination of Ontology, MSO, and prompt instructions within this context. Ontology MSO Prompts is used to explain the meaning of a given keyword, word, phrase, or specific detail at the MSO level.
Generative AI > Ontology MSO Property Prompts
The Ontology MSO Property prompts explain the meaning of a given keyword, word, phrase, or specific detail at the MSO's Property level. The LLM Prompt Instruction provides detailed guidelines for each unique combination of Ontology, MSO, prompt instructions, and Property Name within this context.
Generative AI > LLM Prompt Instructions
The Prompt Instructions page includes clear and concise guidance on the actions the model should undertake, directing its behavior and response to meet specific requirements. Depending on the complexity of the task, there may be multiple instructions provided for the same LLM Task.
Generative AI > Ontology MSO Descriptions
The Ontology MSO Description of all the MSOs (Metadata Standard Objects) present in a given ontology, detailing the type of data and key information each MSO contains. This description is used to identify the relevant MSO for a specific user query.
Generative AI > Ontology MSO Priorities
The Ontology MSO Priorities specifies the priority levels assigned to different MSOs in a given ontology, with five possible priority levels: high, medium, low, and none. Based on the settings from the UI, the MSOs are sent for identification according to the selected priority level:
High - Only high-priority MSOs is considered for identification.
Medium - Both medium- and high-priority MSOs are considered.
Low - All MSOs low, medium, and high are considered.
None - No MSOs are used for identification when the priority is set to none.
Never - Even if an MSO matches in the Arango graph, it will not be used or sent for token identification by the LLM.
Generative AI > Ontology MSO Property Descriptions
The Ontology MSO Property Description of all the MSOs (Metadata Standard Objects) present in a given ontology, detailing the type of data and key information each MSO property contains. This description is used to identify the relevant MSO property for a specific user query.
Generative AI > Ontology MSO Property Priorities
The Ontology MSO Property Priorities specifies the priority levels assigned to different MSOs properties in a given ontology, with five possible priority levels: high, medium, low, and none. Based on the settings from the UI, the MSOs are sent for identification according to the selected priority level:
High - Only high-priority MSO properties are considered for identification.
Medium - Both medium- and high-priority MSOs are considered.
Low - All MSO properties low, medium, and high are considered.
None - No MSOs are used for identification when the priority is set to none.
Never - Even if an MSO property matches in the Arango graph, it will not be used or sent for token identification by the LLM.
Generative AI > LLM Examples
The LLM Examples page includes a few examples based on the given LLM Task.
Each example includes the following:
Natural Language Text (Example User Query)
Response (Example Response)
Reasoning (Example Reasoning for the given query)
Tags (Conditions determining the selection of the example)
Generative AI > LLM Example Documents
The LLM Examples Document page includes the connection between LLM Tasks, LLM Examples, and sections within the LLM Example Document. Depending on an LLM Task and user query, the appropriate LLM Example is chosen, utilizing relevant external knowledge from the corresponding section in the LLM Example Document.
Generative AI > LLM Example Sections
The LLM Example Document Section page includes the precise external knowledge needed for each LLM Example. For example, if a Text2SQL instance necessitates a particular schema or details regarding MSO & MSO Properties, it connects to the relevant section in the LLM Example Document through the Associated Section property within the LLM Example.
Generative AI > Prompt Examples
The Prompt Examples page establishes a correlation between Prompt Templates and LLM Examples. Depending on the chosen Prompt Template associated with the LLM Tool, relevant examples are filtered accordingly.
Generative AI > LLM Function Prompts
The LLM Function Prompts page defines custom functions required for an LLM Task to generate the output response effectively.
Example:
ao_lower: Converts the casing of a varchar or text data type field (column of a table from schema) to lowercase. It accepts one parameter $1: the field to be converted to lowercase. This function applies to:
Fields involved in spatial where criteria, such as fields storing information about county, city, state, country, address, etc.
Generative AI > LLM Models
The LLM Models page includes the master records for the LLM Models referenced by the different LLM Configurations available.
Generative AI > LLM Personas
The LLM Personas page involves assigning a distinct role or identity to the language model in prompt engineering. This influences the style, tone, and content of its responses. By defining a persona, the model can simulate an expert in a specialized field, embody a character with specific traits, or act as a professional possessing targeted knowledge.
Generative AI > LLM Tools
The LLM tools page includes tools designed for specific tasks related to LLM. Each tool is dedicated to a particular function, such as Text2SQL, Document Summarization, or Textual Insights Tool, customized to its unique task.
Generative AI > Reserved Key Words
The Reserved Keywords page lists words, phrases, or tokens with predefined meanings in AI models, frameworks, or programming environments. These keywords help regulate model behavior, enforce restrictions, or define special functions.
Generative AI > Question Generation Instruction Mapping
…
Generative AI > Insight Configurations
…