An Agent-Verified System for Resource-Constrained Query-Focused Summarisation using Small Language Models
Abstract
Abstract Retrieval-Augmented Generation (RAG) powered by Large Language Models (LLMs) has become a prominent paradigm for knowledge-intensive natural language processing, substantially improving the factual accuracy of query-focused summarisation. However, deploying these architectures on resource-constrained edge devices remains highly challenging due to the severe latency and computational overhead of large-scale models. Small Language Models (SLMs) offer a computationally feasible alternative, yet they are considerably more prone to hallucinations and the ``Relevant but Unfaithful'' contradiction. To address this, an Agent-Verified, parallel RAG architecture specifically designed for edge computing is presented in this work. The architecture creates a dual-path verification loop by interrogating both an external Wikipedia agent and a local vector database simultaneously. An SLM Judge dynamically cross-references the retrieved evidence to resolve knowledge disputes mathematically before text generation. Furthermore, this work demonstrates that abstractive fact-checking systems are fundamentally penalised by conventional n-gram evaluation techniques, and presents empirical evidence to support this finding.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.