LLM-Assisted Workflow Graph Generation for Task Mining

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional task mining is time-consuming and requires significant manual intervention, such as manually recording and merging workflows, which hinders efficient workflow analysis and optimization.

Innovation Solution

Utilizing large language models (LLMs) to automatically generate and modify structured definitions of workflows based on natural language descriptions and API mappings, reducing the need for manual recording and merging processes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If conventional task mining is performed by manually recording user interactions, then workflow data can be captured, but the process becomes time-consuming and requires significant manual intervention

Engineering Contradiction:
Improveworkflow capture automationVSAvoidtime for workflow recording
Core Design Contradiction:
Extent of automationVSLoss of time

Solution Approach 1:

The patent replaces manual mechanical recording operations with an automated system that uses machine learning models to capture and analyze workflow data. The system automatically records user interactions, screens, and contextual information without requiring manual intervention, thereby substituting the mechanical process of manual recording with an automated computational approach.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service by allowing the workflow capture process to occur automatically without human intervention. The automated recording system captures data, the machine learning model processes and analyzes it, and workflows are generated autonomously, making the entire process self-sufficient and eliminating the need for manual task mining operations.

Inventive Principle:
Principle #25Self-service

2Productivity

If conventional task mining requires manual merging of workflow records, then workflow consolidation can be achieved, but the process becomes complex and time-consuming

Engineering Contradiction:
Improveworkflow analysis efficiencyVSAvoidmanual intervention complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent replaces manual workflow merging operations with automated machine learning-based consolidation. The system uses trained models to automatically merge and de-duplicate workflow records, eliminating the need for manual merging operations and reducing the complexity associated with manual intervention in workflow consolidation.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The machine learning model acts as an intermediary between raw workflow data and consolidated workflow processes. It automatically processes, analyzes, and merges workflow records, serving as a mediator that handles the complexity of workflow consolidation without requiring direct manual intervention.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If manual workflow recording is performed for every variation, then complete workflow coverage can be achieved, but the time and effort required increases significantly

Engineering Contradiction:
Improveworkflow data completenessVSAvoidtime for recording variations
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by automatically capturing and recording all workflow variations as they occur in real-time without requiring manual intervention for each variation. The automated recording system proactively collects data across multiple workflow scenarios, ensuring comprehensive coverage before analysis begins.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The machine learning model uses feedback from analyzed workflow data to automatically identify and capture additional workflow variations. As the system processes recorded data, it learns from patterns and automatically adjusts to ensure complete workflow coverage, using the feedback loop to maintain reliability while reducing manual recording requirements.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4645067A1Assisted task mining using large language models
Publication Date: 2025.11.05 UIPATH INC
  • EP4645067A1 patent drawingFigure 1
  • EP4645067A1 patent drawingFigure 2
  • EP4645067A1 patent drawingFigure 3

AI summary

Systems and methods for generating a structured definition of a graph of a workflow are provided. 1) a natural language description of a workflow and 2) a description of one or more APIs (application programming interfaces) for performing the workflow are received. A structured definition of a graph of the workflow is generated using a language model based on the natural language description of the workflow and the description of the one or more APIs. One or more nodes in the structured definition of the graph of the workflow is mapped to at least one of the one or more APIs. The structured definition of the graph of the workflow is output.