Leveraging Multi-Agent System (MAS) and Fine-Tuned Small Language Models (SLMs) for Automated Telecom Network Troubleshooting

2026-07-13 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Telecommunications Network Operations · Depth: Expert, extended

Summary

A Multi-Agent System (MAS) is proposed for fully automated telecom network troubleshooting, integrating a Large Language Model (LLM) for orchestration with a fine-tuned Small Language Model (SLM) acting as a solution planner. This system dynamically activates specialized agents, including an orchestrator, solution planner, executor, data retriever, and root-cause analyzer, to diagnose issues and recommend remediation strategies. A key innovation involves fine-tuning an SLM, specifically Unsloth/DeepSeek-R1-Qwen-3-8B, on proprietary troubleshooting documents using Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) with the Transformers Reinforcement Learning (TRL) framework. Experimental results demonstrate that this MAS significantly accelerates troubleshooting automation across both Radio Access Network (RAN) and Core network domains, achieving a six-fold reduction in mean troubleshooting time per node and a 10% improvement in accuracy compared to human-driven processes. The fine-tuned 8B model consistently produced high-quality, domain-grounded solutions, comparable to GPT-4o-mini.

Key takeaway

For MLOps Engineers deploying AI in telecom network operations, this Multi-Agent System provides a robust framework to automate troubleshooting. You should consider fine-tuning Small Language Models on your proprietary documentation to create domain-grounded solution planners. This approach reduces reliance on costly external LLMs, enhances data privacy, and demonstrably cuts mean troubleshooting time by 6x while improving accuracy by 10% across RAN and Core networks, making your operations more efficient and scalable.

Key insights

Combining LLM orchestration with fine-tuned SLMs automates telecom troubleshooting, enhancing efficiency and accuracy.

Principles

• Agentic workflows improve complex task automation.
• Fine-tuning SLMs reduces LLM operational costs.
• Domain-specific RFT enhances model grounding.

Method

The proposed MAS uses an LLM orchestrator to coordinate agents (planner, executor, retriever, RCA) in a ReAct loop. A SLM is fine-tuned via SFT and RFT using TRL on proprietary documents for solution planning.

In practice

• Use HippoRAG for knowledge graph integration.
• Implement uniform chunking for RFT efficiency.
• Design RAGAS-based rewards for domain relevance.

Topics

• Multi-Agent Systems
• Small Language Models
• Telecom Network Troubleshooting
• Reinforcement Fine-Tuning
• RAN & Core Networks
• HippoRAG

Code references

• amun-ai/hypha

Best for: AI Architect, Machine Learning Engineer, NLP Engineer, AI Scientist, AI Engineer, MLOps Engineer

Related on AIssential

• Rethinking Scale: Deployment Trade-offs of Small Language Models under Agent Paradigms — Agent-centric design significantly enhances Small Language Model performance and efficiency in resource-constrained financial applications.
• Evaluating Agentic Configuration Repair for Computer Networks — Agentic LLM architectures significantly improve network configuration repair efficacy and safety through iterative validation.
• FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents — Autonomous agents can significantly automate end-to-end LLM fine-tuning by iteratively refining strategies based on evaluation feedback.
• 5 Ways Small Language Models Are Powering Next-Gen Agents — SLMs are increasingly vital for agentic AI, offering efficiency, cost savings, and on-device capabilities for specialized, repetitive tasks.
• Scaling Meta's Multi-Agent Systems to a Billion Videos — Specialized multi-agent systems with smaller LLMs effectively solve complex, high-volume multimodal problems at scale.
• Small, Free, and Effective: Orchestrating Open-Weight Small Language Models to Outperform Single LLM for Malware Analysis — Orchestrated SLM ensembles, especially with evidence grounding, can outperform single LLMs for specialized tasks like malware analysis.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.