IP Library Granted Patent US 12705437
Granted Patent B2
US 12705437 · App. 18/388,003 · Granted Aug 11, 2026

Evaluation framework for LLM-based network troubleshooting and monitoring agents

Inventors: Eduard Schornig (Haarlem, NL); Jean-Philippe Vasseur (Combloux, FR); Grégory Mermoud (Venthône, CH); Pierre-André Savalle (Rueil-Malmaison, FR)
Assignee: Cisco Technology, Inc.
G06F40/58G06F40/51G06N3/0455H04L41/046H04L41/16
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12705437
App. No.
18/388,003
Granted
Aug 11, 2026
Kind
B2
Abstract

In one implementation, a device uses a large language model-based agent to perform a task in a network, to obtain a first result. The device executes code in the network to perform the task, to obtain a second result. The device makes a comparison between the first result and the second result. The device provides, based in part on the comparison, one or more performance metrics for the large language model-based agent for display.

Claims (49)

1 . A method comprising:

using, by a device, a large language model-based agent to perform a task in a network, to obtain a first result related to a first detected state of the network;

causing execution of predetermined code to perform the task, to obtain a second result related to a second detected state of the network, the execution of the predetermined code including communicating with one or more network devices in the network, the predetermined code being associated with input previously provided by one or more users;

making, by the device, a comparison between the first result and the second result; and

providing, by the device and based in part on the comparison, one or more performance metrics for the large language model-based agent for display.

2 . The method as in claim 1 , wherein using the large language model-based agent to perform the task in the network comprises:

instructing the large language model-based agent to use a particular large language model to perform the task.

3 . The method as in claim 1 , wherein using the large language model-based agent to perform the task in the network comprises:

configuring the large language model-based agent using a parameter that controls at least one of: a prompting strategy used by the large language model-based agent to perform the task or a degree of randomness of a large language model used by the large language model-based agent to perform the task.

4 . The method as in claim 1 , wherein using the large language model-based agent to perform the task in the network comprises:

configuring the large language model-based agent using a parameter that controls at least one of: a maximum number of tokens the large language model-based agent can use to perform the task or a maximum number of steps the large language model-based agent is allowed to perform to complete the task.

5 . The method as in claim 1 , further comprising:

sending, by the device, a request for a subject matter expert to answer a questionnaire; and

translating, by the device, answers to the questionnaire into the task and the predetermined code.

6 . The method as in claim 1 , wherein the task comprises at least one of: troubleshooting an issue in the network, monitoring the network, or configuring a portion of the network.

7 . The method as in claim 1 , wherein the task has an associated difficulty rating specified by a user.

8 . The method as in claim 1 , wherein using the large language model-based agent to perform the task in the network comprises:

sending a text-based question to the large language model-based agent.

9 . The method as in claim 1 , further comprising:

using reinforcement learning to train the large language model-based agent, based on the comparison.

10 . The method as in claim 1 , wherein the one or more performance metrics comprise at least one of: an efficacy metric, a recall metric, a precision metric, an error rate, or a response time.

11 . An apparatus, comprising:

one or more network interfaces;

a processor coupled to the one or more network interfaces; and

a memory configured to store instructions that, when executed by the processor, configure the processor to:

use a large language model-based agent to perform a task in a network, to obtain a first result related to a first detected state of the network;

cause execution of predetermined code to perform the task, to obtain a second result related to a second detected state of the network, the execution of the predetermined code including communicating with one or more network devices in the network, the predetermined code being associated with input previously provided by one or more users;

make a comparison between the first result and the second result; and

provide, based in part on the comparison, one or more performance metrics for the large language model-based agent for display.

12 . The apparatus as in claim 11 , wherein the processor is configured to use the large language model-based agent to perform the task in the network by:

instructing the large language model-based agent to use a particular large language model to perform the task.

13 . The apparatus as in claim 11 , wherein the processor is further configured to:

configure the large language model-based agent using a parameter that controls at least one of: a prompting strategy used by the large language model-based agent to perform the task or a degree of randomness of a large language model used by the large language model-based agent to perform the task.

14 . The apparatus as in claim 11 , wherein the processor is further configured to:

configure the large language model-based agent using a parameter that controls at least one of: a maximum number of tokens the large language model-based agent can use to perform the task or a maximum number of steps the large language model-based agent is allowed to perform to complete the task.

15 . The apparatus as in claim 11 , wherein the processor is further configured to:

send a request for a subject matter expert to answer a questionnaire; and

translate answers to the questionnaire into the task and the predetermined code.

16 . The apparatus as in claim 11 , wherein the task comprises at least one of: troubleshooting an issue in the network, monitoring the network, or configuring a portion of the network.

17 . The apparatus as in claim 11 , wherein the task has an associated difficulty rating specified by a user.

18 . The apparatus as in claim 11 , wherein the processor is configured to use the large language model-based agent to perform the task in the network by:

sending a text-based question to the large language model-based agent.

19 . The apparatus as in claim 11 , wherein the processor is further configured to:

use reinforcement learning to train the large language model-based agent, based on the comparison.

20 . A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising:

using, by the device, a large language model-based agent to perform a task in a network, to obtain a first result related to a first detected state of the network;

causing execution of predetermined code to perform the task, to obtain a second result related to a second detected state of the network, the execution of the predetermined code including communicating with one or more network devices in the network, the predetermined code being associated with input previously provided by one or more users;

making, by the device, a comparison between the first result and the second result; and

providing, by the device and based in part on the comparison, one or more performance metrics for the large language model-based agent for display.