Page 298 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

• REQ-04: It is critical that the processed datasets are available for training and inference,
with selectable training/testing sets and access to a shared compute resource pool for
model development and evaluation.
• REQ-05: It is expected that the platform provides comprehensive model lifecycle
management, including versioning, fine-tuning, deployment, rollback, and real-time
monitoring across training, staging, and production environments.
• REQ-06: It is expected that the agent module supports construction and orchestration of
intelligent agents using tools such as prompt design, model selection, plug-in integration,
API linking, and knowledge base access.
• REQ-07: It is of added value that users can combine trained models with workflow
configurations and knowledge sources to deploy agents tailored to their needs and link
them to external business systems.
• REQ-08: It is critical that all services, including Model-as-a-Service (MaaS) and agent
orchestration, are exposed via secure, scalable, and permission-controlled application
programming interfaces (APIs) for external access and automation.

4 Sequence Diagram

5 References

[1] [ITU_T F.AIAP] Recommendation ITU-T F.AIAP(2025), Framework and requirements for
AI agents platform. Available online: https:// www .itu .int/ md/ T25 -SG21 -250113 -TD -WP3
-0041/ en
[2] [ITU-T F.RF-AIAC-FM] Recommendation ITU-T F.RF-AIAC-FM(2025), Requirements and
Framework for AI Agent Collaboration for Foundation Models. Available online: https://
www .itu .int/ md/ T25 -SG21 -250113 -TD -WP3 -0054/ en
[3] [ITU-T M.3080] Recommendation ITU-T M.3080 (2021), Framework of AI enhanced
Telecom Operation and Management (AITOM). Available online: https:// www .itu .int/ rec/
T -REC -M .3080 -202102 -I

262