Invention Title:

DIGITAL ASSISTANT USING GENERATIVE ARTIFICIAL INTELLIGENCE

Publication number:

US20250094725

Publication date:

2025-03-20

Section:

Physics

Class:

G06F40/35

Inventors:

Vishal Vishnoi Redwood City, CA, United States

Ying Xu Albion, Australia

Xin Xu San Jose, CA, United States

Mark Edward Johnson Sydney, Australia

Aashna Devang Kanuga Foster City, CA, United States

Thanh Long Duong Melbourne, Australia

Srinivasa Phani Kumar Gadde Fremont, CA, United States

Diego Andres Cornejo Barra Chicago, IL, United States

Yakupitiyage Don Thanuja Samodhve Dharmasiri Melbourne, Australia

Assignee:

ORACLE INTERNATIONAL CORPORATION Redwood Shores, CA, United States

Applicant:

Oracle International Corporation Redwood Shores, CA, United States

Smart overview of the Invention

Generative artificial intelligence is leveraged to enhance digital assistants, enabling them to perform complex tasks through natural language interactions. The process begins with constructing an input prompt that includes a user's natural language utterance. This prompt is processed by a generative AI model to create an execution plan, which outlines the actions to be taken by various agents. The execution plan is then executed, and the results are used to generate a response to the user's initial query.

Execution Plan Creation

The creation of an execution plan involves determining suitable agents and their associated actions based on the input prompt. This involves semantic searches on available agents and actions, allowing the system to identify candidates that can fulfill the user's request. The execution plan is structured as an ordered list of actions, ensuring that tasks are performed efficiently and in the correct sequence.

Execution and Response Generation

Upon generating the execution plan, designated agents carry out the specified actions. These actions may require contextual information, which is retrieved as needed. The system then uses a second generative AI model to synthesize a response based on the outputs of these actions. This response is communicated back to the user, completing the interaction.

Handling Complex Interactions

The system is designed to manage complex interactions by considering conversation history and dependencies between actions. It can handle both serial and parallel processing of tasks, adapting to the presence or absence of dependencies. This flexibility enhances its ability to deliver coherent and contextually appropriate responses across a range of scenarios.

System Architecture

The described techniques are implemented via a system comprising processors and computer-readable media that store necessary instructions. These components work together to execute the methods outlined, providing a robust framework for deploying digital assistants powered by generative AI. This architecture supports diverse implementations, allowing for broad application across different domains.