arXiv paper proposes LM-DSB lightweight multi-agent system to automate daily construction safety bulletins

[arXiv] A preprint paper proposes the Lightweight Multi-Agent DSB System (LM-DSB), which combines retrieval-augmented generation with a locally deployable large language model on consumer-grade personal computers to automate daily safety bulletin (DSB) generation in construction. According to the paper, manual DSB creation takes 2-3 hours daily and yields inconsistent quality, with 25% of crews never conducting pretask planning. The system coordinates planner, researcher and writer agents in a hierarchical architecture, with domain-specific fine-tuning on construction safety documentation and retrieval grounding in 37 health, safety and environment (HSE) documents. In a first-phase comparative evaluation, the multi-agent RAG configuration using Llama 3-8B outperformed alternatives, with TF-IDF of 0.70, BLEU of 0.39, ROUGE-1 of 0.66 and a length ratio of 1.01 versus 0.39 for single-agent RAG. A second-phase expert evaluation (N=40) gave mean scores of 3.50-4.20 on a 5-point scale, including 4.20 for regulatory compliance and 4.10 for factual accuracy. The paper cites limitations including knowledge base scope, language constraints and lack of field deployment validation.