Multimodal Action Orchestration Using Moderated LLM Intent Mapping

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional systems process multi-modality inputs inefficiently, leading to fragmented processing and limited understanding of user intent, inflexible mapping of intents to actions, and static user interfaces that fail to consider specific context, resulting in a suboptimal user experience.

Innovation Solution

A unified schema is introduced to standardize user inputs and actions, leveraging LLMs and moderated prompts for accurate intent mapping, and adaptive user interfaces that dynamically adjust based on action requirements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional systems process different input modalities separately using specialized modules, then each modality can be handled with dedicated algorithms, but the processing pipeline becomes fragmented and inefficient

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidprocessing pipeline complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges multiple specialized processing modules into a unified processing pipeline that handles voice, image, and text inputs through a common architecture. The system integrates modality-specific preprocessing with a unified intent mapping and action selection mechanism, eliminating the need for separate specialized modules for each modality while maintaining processing accuracy.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system implements a universal processing framework that can accommodate multiple input modalities through a single standardized interface. The unified schema and moderation rules apply across all modalities, allowing the same core processing logic to handle diverse input types without requiring separate dedicated algorithms for each modality.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If predefined rules or hard-coded mappings are used to determine actions, then the system logic is simple and maintainable, but the system becomes inflexible and fails to adapt to diverse user needs

Engineering Contradiction:
Improveadaptability to user needsVSAvoidsystem logic complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system replaces static predefined rules with a dynamic intent mapping mechanism that uses LLMs to interpret user inputs. The moderation rules and influenced examples allow the system to adapt its behavior based on the specific input modality and context, enabling flexible action selection without hard-coded mappings while maintaining manageable system logic through structured prompt engineering.

Inventive Principle:
Principle #15Dynamics

3Ease of operation

If static user interfaces are provided for all actions, then the interface design is simple and consistent, but the interfaces become cluttered and confusing when specific action requirements are not considered

Engineering Contradiction:
Improveuser interface usabilityVSAvoidinterface customization complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The system implements dynamic user interface generation that adapts the interface elements and layout based on the specific action being performed and the input modality used. The interface dynamically adjusts to present relevant controls and information for each action type, eliminating clutter and confusion while maintaining consistency through a unified design language and structured interface generation process.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4672079A1Orchestrating dynamic actions through multi-modality inputs
Publication Date: 2025.12.31 INTUIT INC
  • EP4672079A1 patent drawingFigure 1
  • EP4672079A1 patent drawingFigure 2A
  • EP4672079A1 patent drawingFigure 2B

AI summary

Certain aspects of the disclosure provide for processing multi-modality inputs to trigger dynamic actions. In examples, a method may include: receiving user input through one or more input modalities; converting the user input and one or more supported actions obtained from an actions repository into a unified schema; applying a set of moderation rules to the unified schema to generate a moderated input; generating a prompt based on the moderated input and one or more influencing examples; processing the prompt using one or more Large Language Models (LLMs) to obtain one or more matched actions and populated data models; augmenting the one or more matched actions with additional data; initiating one or more supported actions based on the one or more matched actions and populated data models; and generating a response to the user input based on results of executing the one or more supported actions.