Uploaded June 2025 | Updated September 2026, 1 week ago
In order to maintain a large-scale network that continues to expand with a limited number of operators, we are working on improving productivity, including network automation, and we have high expectations for the use of generative AI technology.
At present, we are exploring the expansion of our current network automation capabilities to address more advanced use cases, enabling the full delegation of operator tasks.
In this presentation, we would like to discuss the following technical challenges and approaches in network automation using generative AI (Large Language Model agent) about NW trouble shooting, which is a advanced use case for network automation.
■Multi-agent configuration
In operations such as NW trouble shooting, it is not possible to make a decision based on simply repeating questions and answers or to cover with static scripts. Rather, it is necessary to make autonomous outcomes, such as planning multiple investigation procedures and modifying the plan in the middle depending on the outcomes. What makes such decisions possible is a multi-agent configuration in which multiple LLM agents work together, and this presentation will introduce an example of an agent configuration for NW trouble shooting
■Improving the accuracy of operational actions
LLMs such as GPT4o have a wide range of knowledge and their response accuracy is improving day by day, but it seems that they are not able to cover specifications for NW devices.
In response to this, we will introduce in-context learning approaches to improve accuracy.
Speaker Ryosuke Sato - NTT FIELDTECHNO CORPORATION
In order to maintain a large-scale network that continues to expand with a limited number of operators, we are working on improving productivity, including network automation, and we have high expectations for the use of generative AI technology.
At present, we are exploring the expansion of our current network automation capabilities to address more advanced use cases, enabling the full delegation of operator tasks.
In this presentation, we would like to discuss the following technical challenges and approaches in network automation using generative AI (Large Language Model agent) about NW trouble shooting, which is a advanced use case for network automation.
■Multi-agent configuration
In operations such as NW trouble shooting, it is not possible to make a decision based on simply repeating questions and answers or to cover with static scripts. Rather, it is necessary to make autonomous outcomes, such as planning multiple investigation procedures and modifying the plan in the middle depending on the outcomes. What makes such decisions possible is a multi-agent configuration in which multiple LLM agents work together, and this presentation will introduce an example of an agent configuration for NW trouble shooting
■Improving the accuracy of operational actions
LLMs such as GPT4o have a wide range of knowledge and their response accuracy is improving day by day, but it seems that they are not able to cover specifications for NW devices.
In response to this, we will introduce in-context learning approaches to improve accuracy.
Speaker Ryosuke Sato - NTT FIELDTECHNO CORPORATION
