Conversational network assurance using large language models
In one embodiment, a device receives, at a first large language model executed by a device, textual input from a user of a network regarding a networking issue in the network. The device issues, by the first large language model and to a second large language model, one or more questions regarding the network based on the textual input. The device receives, at the first large language model and from the second large language model, one or more answers to the one or more questions. The device generates, by the first large language model, a textual response to the textual input for presentation to the user.
1 . A method comprising:
receiving, at a first large language model executed by a device, textual input from a user of a network regarding a networking issue in the network, the first large language model being trained based on previous conversations with users regarding network issues;
generating, based on the textual input, one or more natural language questions regarding the network;
issuing, by the first large language model and to a second large language model distinct from the first large language model, the one or more natural language questions, wherein the second large language model is trained to semantically parse the one or more natural language questions and generate one or more corresponding structured database queries;
receiving, at the first large language model and from the second large language model, one or more answers to the one or more natural language questions, the one or more answers being generated based on one or more database responses, to the one or more structured database queries, that relate to telemetry data collected from the network; and
generating, by the first large language model and based on the one or more answers, a textual response to the textual input for presentation to the user, the textual response including a response regarding the networking issue.
2 . The method as in claim 1 , wherein the first large language model acts as an intermediate chat model between the user and one or more network support personnel.
3 . The method as in claim 2 , wherein the textual response includes a request that the user confirm validity of information included in the textual response based on the one or more answers, and wherein the method further comprises:
connecting, by the first large language model, the user to the one or more network support personnel when the user does not confirm the validity of the information.
4 . The method as in claim 1 , wherein the one or more natural language questions query at least one of: a location of an endpoint in the network operated by the user or an application accessed by the user via the network.
5 . The method as in claim 1 , further comprising:
initiating, by the first large language model, a chat session with the user, in response to an alert received by the first large language model from a network assurance service for the network.
6 . The method as in claim 1 , wherein the first large language model is trained in part by asking end users to rank or label sample textual outputs of the first large language model.
7 . The method as in claim 1 , wherein the first large language model is trained to interact differently with end users and network support personnel.
8 . The method as in claim 1 , further comprising:
providing, by the device, the one or more natural language questions and the one or more answers for display.
9 . The method of claim 1 , wherein the second large language model is trained to semantically parse the one or more natural language questions and generate one or more corresponding structured database queries according to a known schema.
10 . The method of claim 1 , wherein the one or more corresponding structured database queries comprise SQL queries.
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:
receive, at a first large language model executed by a device, textual input from a user of a network regarding a networking issue in the network, the first large language model being trained based on previous conversations with users regarding network issues;
generate, based on the textual input, one or more natural language questions regarding the network;
issue, by the first large language model and to a second large language model distinct from the first large language model, the one or more natural language questions, wherein the second large language model is trained to semantically parse the one or more natural language questions and generate one or more corresponding structured database queries;
receive, at the first large language model and from the second large language model, one or more answers to the one or more natural language questions, the one or more answers being generated based on one or more database responses, to the one or more structured database queries, that relate to telemetry data collected from the network; and
generate, by the first large language model and based on the one or more answers, a textual response to the textual input for presentation to the user, the textual response including a response regarding the networking issue.
12 . The apparatus as in claim 11 , wherein the first large language model acts as an intermediate chat model between the user and one or more network support personnel.
13 . The apparatus as in claim 12 , wherein the textual response includes a request that the user confirm validity of information included in the textual response based on the one or more answers, and wherein the processor is further configured to:
connect, by the first large language model, the user to the one or more network support personnel when the user does not confirm the validity of the information.
14 . The apparatus as in claim 11 , wherein the one or more natural language questions query at least one of: a location of an endpoint in the network operated by the user or an application accessed by the user via the network.
15 . The apparatus as in claim 11 , wherein the processor is further configured to:
initiate, by the first large language model, a chat session with the user, in response to an alert received by the first large language model from a network assurance service for the network.
16 . The apparatus as in claim 11 , wherein the first large language model is trained in part by asking end users to rank or label sample textual outputs of the first large language model.
17 . The apparatus as in claim 11 , wherein the first large language model is trained to interact differently with end users and network support personnel.
18 . The apparatus as in claim 11 , wherein the second large language model is trained to semantically parse the one or more natural language questions and generate one or more corresponding structured database queries according to a predetermined schema.
19 . The apparatus as in claim 11 , wherein the one or more corresponding structured database queries comprise one or more SQL queries.
20 . A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising:
receiving, at a first large language model executed by the device, textual input from a user of a network regarding a networking issue in the network, the first large language model being trained based on previous conversations with users regarding network issues;
generating, based on the textual input, one or more natural language questions regarding the network;
issuing, by the first large language model and to a second large language model distinct from the first large language model, the one or more natural language questions, wherein the second large language model is configured to semantically parse the one or more natural language questions and generate one or more corresponding structured database queries;
receiving, at the first large language model and from the second large language model, one or more answers to the one or more natural language questions, the one or more answers being based on one or more database responses, to the one or more structured database queries, that relate to telemetry data collected from the network; and
generating, by the first large language model and based on the one or more answers, a textual response to the textual input for presentation to the user, the textual response including a response regarding the networking issue.