Design and Implementation of a Browser-Based Supply Chain Risk Decision Support System Using Large Language Models and Embedding-Based Case Retrieval

Supply chain managers frequently describe disruptions in qualitative language, whereas operational assessment requires explicit assumptions about supply, demand, inventory, and response costs. This paper presents the design and implementation of a Chrome extension that connects these forms of information in a single decision support workflow. The system combines large language model (LLM) extraction, user-confirmed scenario parameters, daily inventory simulation, rule-based assessment, cost-based response allocation, embedding-based case retrieval, and report generation. A side panel supports use alongside existing browser-based work. Operational calculations remain in a deterministic JavaScript engine, while LLMs interpret situations and compose explanations. Case titles, situation descriptions, and responses form retrieval documents; their requested 768-dimensional embeddings are cached locally and compared with a situation-query vector using cosine similarity. Implementation verification uses a supplied automotive-parts dataset and six built-in disruption scenarios. In the pandemic example for a wiring harness, the engine calculates 280 units of unmet demand, with the first occurrence on day 41. A combined response allocates 230 units from inventory transfer and 50 units from additional supplier capacity at an incremental cost of KRW 491,000. These are scenario-model results rather than measured enterprise savings. The contribution is a concrete integration design for exploring LLM and embedding use in enterprise decision support, supported by executable calculations and actual interface captures. Retrieval effectiveness and organizational benefits require subsequent evaluation. The supplementary archive contains only the Chrome extension source code, sample CSV datasets, JSON knowledge base, scenario verification script, reference calculation outputs, and reproduction instructions. Author: Gabkeun Choi, Ph.D. Independent Researcher, Seoul, Republic of Korea. Former Visiting Professor, Department of Media Software, Sungkyul University.