US 2026/0228543 A1Published application

Artificial intelligence agents with domain-specific task adapter layers for large language models

AssigneeQualtrics, LLC

InventorsKalyan Basu, Annamalai Senthilnathan, Kristjan Arumae, Gurdeep Pall, Paul Mayfield

Life of this patent

Estimating the term…

Loading patent…

Abstract

One or more embodiments described herein include an agent adaptation system that uses a centrally hosted agentic model architecture that leverages artificial intelligence models to accomplish tasks to create personalized experiences throughout a user’s product or service journey across multiple domains. Indeed, the agent adaptation system hosts and coordinates a multi-task-specific agent layer framework that receives a user prompt, extracts tasks from the prompts, and utilizes task adapter layers to generate semantic instructions to enable the task-specific agents to complete the tasks.

Claims20

Dependent claims

A computer-implemented method, comprising:

receiving, by an agentic model comprising an orchestrator layer and one or more task-specific agent layers, a user prompt from a client device;
extracting, from the user prompt using the orchestrator layer of the agentic model, one or more tasks executable using a task-specific agent layer of the one or more task-specific agent layers within the agentic model;
generating, by processing the one or more tasks using a large language model augmented by a task adapter layer that is linked to the task-specific agent layer, a semantic instruction comprising data specific to a domain and specific to the one or more tasks and further comprising a command executable by the task-specific agent layer to perform the one or more tasks;
and
performing the one or more tasks by executing, using the task-specific agent layer, the command of the semantic instruction informed by the data specific to the domain and specific to the one or more tasks.

2 more elements

A system comprising:

at least one processor;
and
at least one non-transitory computer-readable storage medium storing instructions that, when executed by the at least one processor, cause the system to:
receive, by an agentic model comprising an orchestrator layer and one or more task-specific agent layers, a user prompt from a client device;
extract, from the user prompt using the orchestrator layer of the agentic model, one or more tasks executable using a task-specific agent layer of the one or more task-specific agent layers within the agentic model;
generate, by processing the one or more tasks using a large language model augmented by a task adapter layer that is linked to the task-specific agent layer, a semantic instruction comprising data specific to a domain and specific to the one or more tasks and further comprising a command executable by the task-specific agent layer to perform the one or more tasks;
and
perform the one or more tasks by executing, using the task-specific agent layer, the command of the semantic instruction informed by the data specific to the domain and specific to the one or more tasks.

5 more elements

A non-transitory computer-readable medium storing instructions thereon that, when executed by at least one processor, cause a computing device to:

receive, by an agentic model comprising an orchestrator layer and one or more task-specific agent layers, a user prompt from a client device;
extract, from the user prompt using the orchestrator layer of the agentic model, one or more tasks executable using a task-specific agent layer of the one or more task-specific agent layers within the agentic model;
generate, by processing the one or more tasks using a large language model augmented by a task adapter layer that is linked to the task-specific agent layer, a semantic instruction comprising data specific to a domain and specific to the one or more tasks and further comprising a command executable by the task-specific agent layer to perform the one or more tasks;
and
perform the one or more tasks by executing, using the task-specific agent layer, the command of the semantic instruction informed by the data specific to the domain and specific to the one or more tasks.

2 more elements

Dependencies are inferred from the claim text.