Retrieval-Augmented Workflow Generation for Accurate Low-Code Automation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital workflow tools require specialized knowledge of workflow programming paradigms and execution paths, posing a barrier for novice users who wish to create functional workflows using natural language descriptions.
Innovation Solution
A retrieval-augmented digital workflow generation technique that uses a neural retriever to select validated components and a generative AI model to generate accurate workflows from natural language descriptions, minimizing hallucinations and resource intensity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a low-code or codeless workflow tool is provided to enable novice users to create workflows using natural language, then ease of operation is improved, but reliability deteriorates due to hallucinations and inaccurate workflow generation
Solution Approach 1:
The patent introduces a retrieval-augmented generation system that acts as an intermediary between the user's natural language input and the workflow generation. The system retrieves relevant information from external knowledge sources and validates components before generation, reducing hallucinations while maintaining ease of use. This mediator layer ensures that generated workflows are both user-friendly and accurate.
Solution Approach 2:
The system implements feedback mechanisms where the generated workflow is validated against retrieved information and external knowledge sources. If hallucinations or inaccuracies are detected, the system can iteratively refine the generation process. This feedback loop ensures reliability while preserving the low-code accessibility for novice users.
2Ease of operation
If a generative AI model is used to create workflows from natural language descriptions, then ease of operation is improved, but use of energy worsens due to high resource intensity
Solution Approach 1:
The patent segments the workflow generation process into distinct stages: natural language input processing, information retrieval from external sources, validation against retrieved data, and final generation. By dividing the task, the system can optimize each stage separately, reducing overall computational resource consumption while maintaining ease of operation for users.
Solution Approach 2:
The system performs preliminary actions by retrieving relevant information and validating potential workflow components before the actual generation process. This pre-processing step reduces the computational burden on the generative model during the main generation phase, lowering energy consumption while preserving natural language input capabilities.
3Reliability
If validation mechanisms are added to minimize hallucinations in generated workflows, then reliability is improved, but device complexity worsens
Solution Approach 1:
The patent uses an intermediary retrieval-augmented generation architecture that integrates validation mechanisms without significantly increasing overall system complexity. The retrieval component serves as a mediator that provides validated information to the generation process, ensuring reliability while maintaining a manageable system structure through modular design.
Solution Approach 2:
The system employs multi-functional components that perform both retrieval and validation functions within a unified framework. This universal approach allows the same infrastructure to serve multiple purposes (information retrieval, validation, and generation support), improving reliability without proportionally increasing device complexity.
Data Source
AI summary
A description of a desired computer workflow is obtained. Based on the description of the desired computer workflow, a set of prospective components from a plurality of components is selected. A prompt is generated using the description of the desired computer workflow and the set of prospective components. Based on providing the prompt to a large-language-model, at least a portion of code associated with the desired computer workflow is generated.