Authors:
- Fatih E. NAR, Distinguished Chief Architect at Red Hat & LFN AI TAC member
- Ranny Haiby, CTO Networking, Edge, and Access, LFN
The latest technology advances in AI bring new opportunities to many industry verticals. For the wireless network operators it means shifting from traditional, reactive methods to a proactive, AI-driven approach centered around user satisfaction. In a recent Medium article our community members delved into the evolution of 5G network management, particularly focusing on how artificial intelligence (AI) is transforming the way networks are operated and optimized. The core concept revolves around understanding and predicting user needs to enhance their experience.
The article outlines a framework involving three key stages in this evolution:
- Stage 1 – The Outsourcing Era – was characterized by heavy reliance on external AI services, leading to limited control, high costs, and constrained innovation.
- Stage 2 – The Independence Movement – saw providers adopting open-source models and building internal expertise for greater autonomy, customization, and cost control.
- Stage 3 – The Practical Reality Check – represents the current state, focusing on a balanced approach with hybrid deployment models, operational integration, performance optimization, and the development of governance frameworks.
A plan for AI implementation
The article proposes 5 step plan for the Telco AI transformation:
Level 1: Democratize Knowledge Access . Leverage existing internal knowledge bases (products documents, source code repositories, support ticket archives etc) with key-word search capabilities to enable information retrieval for technical support and troubleshooting.
Level 2: On-Premise AI Deployment. Implement in-house AI models (like LLaMA, Mistral, DeepSeek) with efficient serving technologies (like vLLM) for data sovereignty, customization, and predictable performance; either from model training scratch and/or fine tuning of an out-of-shelf model from public model repositories (such as Hugging Face).
Level 3: Retrieval Augmented Generation (RAG) with Semantic Retrieval. Move beyond keyword search to understand the meaning of queries using embedding models and vector databases, enabling better intent recognition and cross-domain knowledge access and data-meshing.
Level 4: Intelligent Context. Analyze queries to dynamically assemble the most relevant information from various sources, optimizing the context provided to AI models for complex problem-solving.
Level 5: Autonomous Agents. Evolve towards AI agents that can independently execute tasks within defined boundaries, roles and responsibilities, utilizing protocols like Model Context Protocol (MCP) and Agent Communication Protocol (ACP) for performant and scalable solution architectures for real-time problem identification and resolution, which leads us to true Autonomous Networks.
Avoiding re-invention of the wheel – the network domain-specific AI framework
The Essedum open source project, recently announced by the Linux Foundation, is an AI application development framework under the Linux Foundation Networking (LFN). Essedum aims to accelerate the integration of AI data, models, and applications within the networking industry. Built upon Infosys’ existing AI networking solutions, it seeks to provide a unified framework for developing domain-specific AI tools, ultimately fostering smarter and more efficient network operations by making AI more accessible and manageable for telecom operators.
Essedum provides a reusable and modular framework aimed at accelerating common tasks involved in AI Application and Agent development for networking use cases. That includes Governance tools that ensure proper use of AI and trustworthiness of the agent operation, Accelerators that help with integration of data and models, and MLOps tools that are required for the execution of the agents and applications. To find out more about Essedum please visit https://essedum.org/
