Page 296 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Pretrained on open and proprietary corpora, then fine-tuned on multimodal
Model Training and telecom data. Human-in-the-loop supervision ensures task alignment. Uses
Fine-Tuning parameter-efficient tuning, reinforcement learning, and distributed training
frameworks. Model performance is tracked via telecom KPIs.
we use pilot deployment from
Testbeds or Pilot http:// 172 .27 .221 .54: 3000/ agent -platform -web/ login ,
Deployments
but do some processing in the combination with telecom data features.

Code repositories Inside China Telecom

2 Use Case Description

2�1 Description

As the scale and complexity of network operations continue to grow, traditional AI workflows face
challenges such as fragmented development processes, insufficient adaptation to multimodal
data, and limited automation in decision-making. Moreover, network-facing intelligent agents
often lack standardized, secure, and flexible interfaces to interact with operational systems,
limiting their ability to perform real-time diagnostics, control, and service orchestration.

China Telecom has built the industry's entire process toolchain platform that can provide
large model dataset processing, training, application construction, and services. The platform
can provide multi-modal network professional data processing, powerful computing power
support, rich model resources, advanced training and inference methods, and stable training
and inference architecture. At the same time, it opens up secure capability interfaces, allowing
developers to easily access, customize, and expand large model capabilities to meet diverse
needs in different scenarios.

Intelligent agents built through the platform interact with network systems via standardized
internal APIs and orchestration frameworks. These interfaces expose functions such as
alarm monitoring, ticket issuance, resource querying, configuration adjustment, and service
provisioning. By accessing these APIs, agents can autonomously retrieve real-time status,
trigger workflows, or execute operational commands under controlled policies, enabling
closed-loop automation.

This platform can simplify the entire process of large model corpus processing, training,
AI application construction, model deployment and management, greatly improving the
widespread application of AI. The platform builds a comprehensive Model-as-a-Service (MaaS)
system, which include API, intelligent agent and model deployment, providing customized
artificial intelligence solutions more flexibly to meet the needs of different scenarios. Currently,
China Telecom are building a 100+ application ecosystem based on toolchains, covering
various network fields across the entire network, promoting the transformation and upgrading
of cloud network intelligence, and empowering practical scenarios. In the future, we will further
focus on researching cross industry customization, technology integration, etc., to provide users
with more convenient, efficient, and personalized intelligent experiences, while also promoting
the rapid development and innovative upgrading of the entire AI industry.

260