Speakers
Description
Designing reliable, adaptable network services at scale demands more than just automation scripts. It can be achieved with orchestration systems that are modular by design, aligned with industry standards, and capable of evolving across domains. This principle shaped the development of a production-grade orchestration stack for network service provisioning, delivered for the Polish NREN, PIONIER, under the GÉANT GN5-2 project. Built on TM Forum compliant models, dynamic inventory, and workflow automation, the system integrates seamlessly from service design to device configuration, thus demonstrating how structured orchestration can move from architecture to deployment with clarity and control.
The experience has shown that as network orchestration becomes increasingly complex, traditional coding approaches struggle to keep pace with the demand for rapid, yet standards-aligned service design. Thus, our aims extended to exploring the use of Large Language Models (LLMs) as active copilots in the architecture, design, and implementation of full-stack orchestration solutions. Rooted in the idea of vibe coding, we guide the LLM to code every piece such as workflow task, operators for the Source of Truth, as well as schema-compliant payloads and retry-aware logic.
The talk shares insights from both phases: the architectural choices behind the working system, and the experimental use of AI to reproduce or extend parts of it. We discuss practical prompt strategies for generating modular logic, enforcing standards such as the use of TM Forum Open APIs, and guiding newcomers through complex orchestration patterns.
Findings show that while human careful guiding remains critical, prompt-driven workflows can reduce time-to-design and deploy, support maintainability, and offer a viable onboarding path for teams new to network automation. Combined with open APIs and reusable building blocks, this approach enables orchestration that is not only functional, but also teachable, transferable, and ready for broader adoption.
| Estimated presentation duration | 25-30 mins |
|---|
