TelcoAI: Advancing 3GPP Technical Specification Search through Agentic Multi-Modal Retrieval-Augmented Generation

TL;DR

TelcoAI tackles the challenge of querying dense, multi-modal 3GPP specifications by introducing an agentic multi-modal Retrieval-Augmented Generation framework with section-aware chunking, metadata-guided retrieval, and image-text fusion. The system demonstrates strong gains in recall and faithfulness on real-world and benchmark datasets, outperforming state-of-the-art baselines by up to ~16% in recall and achieving around 93% accuracy on the TSpec-LLM benchmark. Ablation studies confirm the value of each architectural component, including hierarchical and hybrid retrieval, post-retrieval filtering, and multi-modal fusion. The work offers a practical pathway to more accessible and reliable 3GPP documentation processing, with potential for extending to multi-turn, cross-version, and broader multi-modal scenarios in telecommunications research and engineering.

Abstract

The 3rd Generation Partnership Project (3GPP) produces complex technical specifications essential to global telecommunications, yet their hierarchical structure, dense formatting, and multi-modal content make them difficult to process. While Large Language Models (LLMs) show promise, existing approaches fall short in handling complex queries, visual information, and document interdependencies. We present TelcoAI, an agentic, multi-modal Retrieval-Augmented Generation (RAG) system tailored for 3GPP documentation. TelcoAI introduces section-aware chunking, structured query planning, metadata-guided retrieval, and multi-modal fusion of text and diagrams. Evaluated on multiple benchmarks-including expert-curated queries-our system achieves $87\%$ recall, $83\%$ claim recall, and $92\%$ faithfulness, representing a $16\%$ improvement over state-of-the-art baselines. These results demonstrate the effectiveness of agentic and multi-modal reasoning in technical document understanding, advancing practical solutions for real-world telecommunications research and engineering.

Figures (1)

• Figure 1: Overview of TelcoAI's architecture showing both ingestion and Q&A pipelines. Below, detailed workflows show the key components of each stage: indexing, pre-retrieval, context retrieval, post-retrieval and answer generation.