Page 218 - AI for Good-Innovate for Impact

AI for Good-Innovate for Impact

realize the full process from scheme generation to task decomposition, logical arrangement,
perception analysis, automatic execution、and optimization feedback.

Advantages of the solution:

(1) Innovation in human-robot interaction mode: Combining ChatOps with LLM technology,
utilizing natural language instruction matching automation capabilities, and using the
intent recognition ability of LLM to extract API input parameters, providing frontline
maintenance personnel with mobile data quick query and execution capabilities.
(2) Innovation of Function Matching Algorithm: In the initialization stage, a multi to one vector
matching relationship is established between natural language samples and several
automation capabilities. Users continuously use natural language to call these capabilities,
continuously increase training samples through iterative feedback, and optimize the
vector matching degree between user input and ChatOps function in real time, achieving
dynamic autonomous learning.
(3) The industry's first network LLM intelligent agent technology: The network LLM provides
intention decomposition, function and parameter analysis, automatic parameter
completion, and real-time orchestration. It forms a dynamic process containing one or
more operation and maintenance actions and drives the intelligent agent to complete
all operations and maintenance actions. The maintenance and execution results are fed
back, iteratively enhancing the intelligent agent's ability.

UN Goals:
• SDG 9: Industry, Innovation, and Infrastructure:

Justify UN Goals selection: SDG 9: Industry, Innovation, and Infrastructure: Enhancing
infrastructure efficiency and innovation in network maintenance.

Promotion idea: Provide pocket-style digital operation portable toolbox to telecommunications,
government, and enterprise customers.

Application effect:

(1) Reducing network maintenance costs: Taking the promotion situation of Jiangsu Telecom
in the province as an example, more than 390 robots of various types were registered, with
434 associated API applications. The daily usage exceeded 800 people and exceeded
2000 times, saving about 2000+ hours/day in query, emergency, and work order disposal
time in Jiangsu Province.
(2) Improving the efficiency of cloud network operation: The expert knowledge base of
network LLM has been constructed for 9 scenarios, based on multi round dialogue,
context parameter extraction, automatic parameter correction and completion to assist
end-to-end task execution and tool calls, with an average accuracy of over 80%. The
real-time orchestration and execution technology of network LLM intelligent agents has
covered 8 specialties in Jiangsu Province. The daily real-time orchestration of Jiangsu
Telecom's provincial management equipment network exceeds 200 times, reducing
the steps of operating equipment or logging into network management queries, saving
20 hours/day of work order processing time, and increasing the work order automation
processing rate by 10%.
(3) Serving over a hundred clients with a production capacity, cultivating a development
oriented operation and maintenance team of over 800 people. Sustainable provision
of convenient and innovative digital self-service channels to the telecommunications,
government and enterprise customers.

202