As the complexity of telecoms networks grows, they are becoming extremely challenging to operate. Running today’s networks typically requires highly skilled staff to conduct a comprehensive analysis of data collected from multiple systems. In some scenarios, it’s becoming almost impossible to manually manage network operations.
At the cutting edge of AI, large language models (LLMs) have yet to be widely deployed to support network operations scenarios. This Catalyst aims to develop an LLM-based AI agent to support future network operations. The solution will be designed to automatically provision access point names (APNs) and data network names (DNNs) based on intent. The aim is to automate 80% of operations and reduce the number of daily man-machine interactions from more than 100 to 10, while cutting service provisioning times from three months to a matter of minutes.
The Catalyst intends to develop a network LLM-based AI agent by consolidating and orchestrating knowledge accumulated in existing systems. The AI agent will then be tested performing tasks that were previously performed by experts. It will employ an LLM to understand user intents, break down and analyze tasks, and then provide the best solutions.
The Catalyst will use a digital twin online simulation to verify the effectiveness of the intent-driven APN/DNN provisioning performed by the agent. It will conduct a before-and-after comparison of key network operations metrics, such as mean-time-to-repair and average handle time (operational efficiency metrics), service availability and other network quality metrics, and the cost of field maintenance.