Page 332 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Awards:

1) " A New Mode and Practice of Human-Robot Interaction Operation and Maintenance
Based on Network Large Language Model‐AI Chat Operations, AIChatOps‐" selected
as a featured case in the ITU AI for Good - Innovate for Impact Final Report 2024.
2) "Intelligent Configuration Auditing" was selected as an industry standard in the Blue Book
of Telecom Industry Development.
3) “Artificial Intelligence for IP Network Configuration Auditing” was selected as one of the
top ten pioneering cases of “Artificial Intelligence + Telecom Industry” organized by
CAICT.
4) "Innovative Applications of China Telecom's Network LLMs in Network Maintenance"
certified as an Application Innovation Case by the China Communications Standards
Association (CCSA).
5) "Cloud-Network Anomaly Response Center - Watcher" selected as a Starlight Forward -
looking Case for Network Intelligence Perception by CAICT 2024.
6) First Prize in the 2024 China Telecom Group Hackathon - LLM Track.

2�3 Future Work

To further enhance this use case, future endeavours include:

• Data Collection: Collect a more comprehensive dataset to improve the model and
enhance the accuracy of anomaly detection and task planning. Optimize the process of
automatic high-quality data collection, enabling the cloud-network operation model to
self - learn and improve over time.
• Model Development: Improve Large Language Models (LLMs) to better understand and
execute complex network maintenance tasks.
• Extension and Variation: Expand the Agentic AI framework to new scenarios, covering
industries other than telecom and government affairs.
• Industry standards: Standardize the full - process solutions of data, models, and tools to
provide a reference for network maintenance across industries.

3 Use Case Requirements

• REQ-01: It is critical to deploy a strategic human-intervention agent framework for
deterministic cloud-network O&M.
Note: The framework allows secondary modification and confirmation of agent intent
orchestration and model parameter extraction, avoiding model hallucinations and
ensuring 100% execution accuracy.
• REQ-02: It is critical to build an automated data collection enhancement algorithm.
Note: The dataset covers full cloud-network scenarios (such as emergency response,
change operations, and alarm monitoring), which can be used to train and fine-tune LLMs,
improving model accuracy to over 95%.
• REQ-03: It is critical to develop a series of automated O&M tools (e.g., configuration
auditing, log auditing) based on large audit models, leveraging real-time network
configurations for incremental fine-tuning of audit LLMs.
• REQ-04: It is critical to achieve end-to-end control of cloud-network O&M emergency
events to ensure timely fault discovery, accurate localization, and well-documented
disposal.

296