Analysis Experience
This article explains the advanced analysis and operational capabilities available in askEdgi. It describes how to validate AI prompts, monitor execution progress, review AI reasoning, perform post-analysis actions, provide feedback on AI-generated responses, and apply approved data cleanup rules.
AI Prompt Validation, Execution Progress & Hover Insights
askEdgi now supports prompt validation with preview mode, execution progress tracking, and hover-based execution summaries to help users validate AI enrichment logic before processing large datasets.
This enhancement improves accuracy, transparency, performance efficiency, and user confidence when running AI enrichment workflows.
Validate Prompt (Preview Mode)
Users can validate AI prompts by executing AI enrichment on a small sample set (default: 5–10 rows).
Functionality
Executes AI enrichment only on sample rows.
Displays preview output within the enrichment modal.
Allows users to edit prompts and re-validate before running full execution.
Supported across Prompt Analysis, Sentiment Analysis, Intent Analysis, and all AI Enrichment functions.
Preview does not modify actual table data.
Execution Control & Progress Tracking
Execution Behavior
Full execution is initiated only after prompt validation (if enforcement is enabled).
UI displays:
AI function name
Execution status
Row-level progress percentage (0–100%)
Live progress updates for multiple sequential enrichment functions
User Guidance
If a user attempts execution without validation, the system displays: “Please validate your prompt before execution.”
Hover Card Details (Post-Execution Function Summary)
After execution, users can hover over the AI function badge to view a detailed summary card containing:
Function type (Prompt Analysis, Sentiment Analysis, etc.)
Input column(s)
Output column
Prompt text (if applicable)
Total rows processed
Success/failure record count
Confidence or AI accuracy score (when available)
Execution timestamp
Processing duration
Purpose
Improves auditability and traceability
Helps users understand how AI outputs were generated
Supports debugging and validation of enrichment logic
Thoughts Feature in askEdgi
askEdgi includes a Reasoning Transparency Panel, accessible through a Thoughts button on AI-generated responses. This feature provides visibility into how the system interprets user prompts, selects relevant data sources, applies policies, and constructs final answers.
The panel is designed to improve transparency, trust, explainability, and troubleshooting while ensuring that sensitive internal reasoning remains secure and sanitized.
Why Reasoning Transparency Matters
Users often need clarity on:
How askEdgi interpreted a prompt
Why certain tables, columns, or glossary terms were selected
Which sources contributed to the answer
How policies and permissions influenced the result
This feature helps:
Reduce misunderstanding of AI responses
Improve prompt refinement and analytical accuracy
Support debugging for internal teams and support users
Build confidence in data-driven insights
Accessing the Thoughts Panel
A Thoughts button appears next to every AI-generated response in askEdgi.
The button:
Appears only for messages generated by the askEdgi AI agent
Does not appear for user messages
Opens a modal overlay when selected
The Thoughts icon visible only within the current active chat and disappears when the user switches to a different chat.
Thoughts Modal Content (MVP)
When opened, the modal displays structured and sanitized reasoning details in multiple sections.
1. Searching and Interpreted Intent
This section explains how askEdgi understood the user request, including:
Interpreted user query
Detected entities
Classified intent such as aggregation, join, or comparison
This helps users understand how their prompt was parsed and interpreted.
2. Source Objects Used
This section lists all internal sources referenced while generating the response, including:
Catalog objects accessed
Glossary terms referenced
Workspace tables and columns used
Lineage paths traversed
Metadata such as classifications, policies, and descriptions
Each source entry displays:
Object name
Object type
Reason it was used
Sources are presented in a structured and organized list.
3. Reasoning Summary
This section provides a high-level summary of how the answer was formed, such as:
Identifying entities and relevant metrics
Matching columns using synonyms or metadata
Applying access controls and governance rules
Determining relevant joins or calculations
The reasoning is sanitized and summarized, ensuring the internal chain of thought is not exposed.
4. Execution Steps and Query Plan (If Applicable)
When relevant, the modal may include:
Steps executed by the AI agent
Query planning or execution context
Applied filters and policy constraints
If query execution fails, the reasoning is displayed up to the failure point, along with a failure message.
Security and Sanitization Principles
The Thoughts panel does not expose the raw internal chain of thought.
Instead:
Reasoning is presented in a safe, structured, and summarized format
Sensitive internal logic remains hidden
Only high-level, explainable insights are displayed
This ensures transparency without compromising system integrity or security.
Post Analysis Actions and Operational Execution
askEdgi supports operational actions that can be executed based on analysis results. These actions allow users to convert analysis output into governed execution within the platform.
The action framework connects analytical results with downstream operations such as Projects, Data Quality, Service Desk, Governance Catalog, tagging, ticket creation, task management, and data quality rule creation, while maintaining control, consistency, and audit tracking.
Configure Action
Configure Action allows users (Recipe Creators) to define what action should be performed on analysis results. It is available from the three-dot menu on result tables and is used to map output columns to the required fields of the selected action.
Once configured, the action becomes part of the recipe and can be executed later through Take Action. If an action is already configured, opening Configure Action directly shows the mapped configuration, allowing updates or removal.
Click the three-dot menu on result tables, alongside Download and Fullscreen options.
Key capabilities
Single Action
Only one action can be configured per output table
Field Mapping
Map output columns to action fields
Mandatory Validation
Required fields must be mapped before saving
Static Values
Default/static values can be assigned
Once configured:
The configuration is linked to the recipe/output
The Take Action option becomes available
The configured action becomes available for execution
The configuration can be updated anytime
Only one action can be active per output table at a time
Removing an action clears all mapped configurations
Configuration is reused during execution without re-mapping
Supported Actions
askEdgi supports multiple operational actions that can be configured on analysis outputs. These actions allow users to directly act on insights without leaving the analysis context.
The supported actions include:
Assign Tag: Apply tags to selected data assets based on analysis results.
Create Data Quality Service Desk Ticket: Raise service desk tickets for identified data quality issues.
Create Data Quality Rule: Define and create data quality rules from analytical outputs.
Create Project Tasks: Convert results into structured tasks within Projects for execution and tracking.
Post Question: Post questions to the Question Wall for further clarification or collaboration.
Each action is designed to convert analysis results into a specific operational outcome within the platform, while maintaining governance and control.
Field Mapping
Field mapping defines how columns from the askEdgi output are linked to the required fields of the selected action. This ensures that the execution uses structured and relevant data from the analysis.
During configuration, users map output columns to action fields. Some fields are mandatory and must be completed before the configuration can be saved or updated. The system validates these fields and enables the Save or Update option only after all required mappings are provided.
Optional fields can be left unmapped if they are not required for execution. In such cases, the system either ignores those fields or applies default handling where applicable.
In addition to dynamic mapping from output columns, users can also define static values. This is useful when a fixed value needs to be applied across all records, such as a constant priority, category, or status.
Once configured, the mapping is saved as part of the recipe. This ensures consistency across executions and allows the same logic to be reused without reconfiguration. Users can revisit and modify the mapping later through the Configure Action flow, and any updates will overwrite the existing configuration.
Unassign Action
Users can remove an existing action configuration.
Steps:
Click Configure Action
Click Unassign Action
Confirmation popup:
Yes: Action removed, action list displayed
No: Return to mapped configuration
Take Action
Take Action enables users to execute configured actions on analysis results directly from the askEdgi output. It acts as the execution layer of the action framework, allowing users to convert analytical results into governed operations within the platform.
The Take Action option is available on the result table. When an action is already configured and the recipe is executed, it is also displayed as a primary button for quick access. If no action is configured, users can still access Take Action through the three-dot menu.
All executions are performed as background jobs to ensure scalability and performance. The system provides execution tracking through Job IDs and displays the final outcome within the chat interface.
Execution Flow
The execution process follows a structured and controlled sequence to ensure accuracy and governance:
The user reviews the analysis output and selects one or more rows from the result table.
The user clicks Take Action from the available options.
The system displays the configured action details along with the mapped fields for the selected records.
The user reviews the data and confirms the execution wherever applicable.
Once confirmed, the system initiates a background job to process the request.
A Job ID is generated and displayed to the user for tracking purposes.
The system processes each selected record based on the configured action logic.
Upon completion, the system displays a detailed execution summary in the chat.
Execution Result Summary
After the job is completed, the system presents a clear and structured summary of the execution outcome. The results are divided into two sections:
Successfully processed records
Failed records
Each section is displayed separately with corresponding data entries, allowing users to easily understand which records were processed and which require further attention.
Action: Create Project Tasks
The Create Project Tasks action allows users to convert analysis results into structured tasks within the Projects module. This helps in operationalizing insights by assigning ownership, tracking progress, and ensuring accountability.
Project Resolution Logic
When the action is executed, the system determines whether to use an existing project or create a new one.
Field Mapping and Requirements
During configuration, users map output columns to task-related fields. Among these:
Task Name and Assignee are mandatory and must be provided for task creation.
Other fields, such as descriptions, object references, and dates, are optional and can be mapped if required.
If optional fields are not provided, the system applies default handling wherever applicable.
Task Type Determination
The system automatically determines the type of task (object or non-object) based on the availability of object-related fields:
If Object Type and Object ID are provided, the task is created as an object-linked task.
If these fields are not provided, the task is created as a non-object task.
The same logic applies to parent tasks using Parent Object fields.
Parent-Child Task Handling
The system supports hierarchical task creation based on the presence of parent task information:
If the Parent Task Name is provided, tasks are grouped under a parent task.
If the Parent Task Name is not provided, tasks are created as standalone entries without any hierarchy.
For grouping:
Parent task names are normalized using trimming and case-insensitive comparison.
If parent object fields are provided, grouping is based on Parent Object ID instead of name.
This ensures consistent grouping and avoids duplicate parent creation.
Additional Execution Behaviors
If the task name is missing in a row, that row is skipped, and processing continues for the remaining records.
If duplicate object tasks are detected, the entire action fails due to validation rules in the Projects module.
If the assignee is inactive or not found, the task is assigned to the Project Owner.
All valid assignees are automatically added as members of the project.
If start and end dates are not provided, tasks inherit dates from the project configuration.
Execution Characteristics
Task creation is executed asynchronously using the Take Action job framework.
Action: Create Data Quality Rule
The Create Data Quality Rule action allows users to convert analysis results into enforceable data quality rules directly within the Data Quality module.
Configuration and Mandatory Fields
To ensure valid rule creation, the following fields must be mapped during configuration:
Object Type
Function (rule logic)
Success Criteria
Rule Name
Rule Purpose
Dimension
The system does not allow saving the configuration unless all mandatory fields are mapped.
Execution Behavior
This action follows a controlled and review-based execution process:
Rules are configured by the recipe creator but executed by the user.
During execution, users must review the rule details for each selected row.
Users explicitly approve the creation before execution begins.
Multiple rules can be processed within a single job.
Execution Flow
The user selects one or more rows from the result table.
The user initiates Take Action and selects the Data Quality Rule action.
The system displays rule details for review.
The user approves the selected entries.
A background job is triggered.
Rules are created in the Data Quality module upon successful execution.
State Management
The review and approval state is tied to a specific execution.
If the recipe is executed again, previous selections and approvals are reset.
Only newly generated results are considered for action.
Audit Trail Tracking
All actions performed through Take Action are recorded in Audit Trails to ensure traceability and governance.
Logging
Each action execution captures detailed information, including:
Action name
Job ID
Execution status (start and completion)
The user who triggered the action
Execution date and time
Module-Level Tracking
Depending on the action performed, logs are recorded in the respective modules:
General Take Action logs are stored under askEdgi in Audit Trails
Service Desk actions are tracked under the Service Desk module
Tag assignments include a source indicator marked as “askEdgi”
Data Quality Rule creation is tracked in both Audit Trails and rule-level history
Term creation is tracked within the Governance Catalog
Only one action can be configured per output table at a time.
All executions follow user permissions and governance controls.
Actions are executed asynchronously to support scalability.
Results are always displayed in the chat with a clear success and failure summary.
Feedback (Thumbs Up / Thumbs Down)
askEdgi allows users to provide feedback on generated responses directly from the chat interface using Thumbs Up and Thumbs Down options. This feedback helps organizations identify response quality, improve metadata curation, and continuously enhance the overall search and analysis experience.
User Feedback Flow
After askEdgi generates a response, users can evaluate the result using the Thumbs Up or Thumbs Down options displayed with the response.
Thumbs Up
Use Thumbs Up when the response is helpful, accurate, and meets the user’s expectations.
When selected:
The response is recorded as positive feedback for the generated response.
No additional input is required.
Thumbs Down
Select Thumbs Down when the response is incorrect, incomplete, irrelevant, or does not meet expectations.
When selected,
A feedback window opens.
User must select what type of issue do you wish to report?* feedback category, such as:
Didn't understand the question
Incorrect answer
Incomplete answer
Missing or incorrect data
Optionally, user can provide additional comments.
Submit the feedback.
After submission, askEdgi displays a confirmation message indicating that the feedback has been successfully recorded.
Example
A user asks:
"Show the top 10 customers by sales."
askEdgi returns a list of customers ranked by sales.
If the result is correct and useful, the user selects Thumbs Up to indicate positive feedback.
If the result is incorrect, incomplete, or does not match the request, the user selects Thumbs Down, chooses a feedback category such as Incorrect Answer, and optionally enters comments like:
"The result shows top 10 products instead of top 10 customers."
The feedback is then recorded and can be used to analyze response quality and identify areas for improvement.
Feedback Processing
When feedback is submitted, askEdgi records:
User question
Generated response
Feedback type
User comments
Associated metadata and assets used during response generation
Feedback Review and Analysis
Feedback is stored within the askEdgi feedback repository(edgi_feedback) and can be reviewed by authorized administrators and product teams. The recorded feedback can be analyzed to identify:
Frequently reported response issues
Missing or incomplete metadata
Incorrect asset recommendations
Semantic mapping gaps
Missing business context
Areas requiring governance improvements
Governance Benefits
Feedback provides a structured mechanism for improving enterprise knowledge and metadata quality. Organizations can use feedback trends to:
Improve table and column descriptions
Refine glossary definitions
Enhance business context
Strengthen semantic mappings
Improve asset discoverability and relevance
These improvements increase the quality of contextual information available during RAG-based retrieval and analysis, leading to more accurate and relevant responses over time.
Feedback is used for governance and metadata improvement activities. Feedback is not used to retrain or fine-tune the underlying AI model.
Data Cleanup
Data Cleanup allows users to apply approved remediation rules directly within the askEdgi workspace without modifying the source data.Data Cleanup allows users to apply approved remediation rules to datasets that contain legacy data quality issues directly within the askEdgi workspace when those issues cannot be corrected in the source system.
For more information, please refer to Data Cleanup.
Copyright © 2026, OvalEdge LLC, Peachtree Corners, GA USA
Was this helpful?