<![CDATA[Small manufacturing and craft-based firms increasingly use Generative Artificial Intelligence (GenAI) through public chat interfaces, low-cost tools, and informal experimentation. However, these firms often make operational decisions with incomplete records, tacit owner knowledge, fragmented spreadsheets, and limited managerial capacity. Under such conditions, open-ended chatbots may generate fluent but unsafe recommendations by overlooking missing information, contradictory evidence, feasibility constraints, and implementation constraints. This study presents Asisten Cerdas Industri Kecil as a bounded Custom GPT artifact for operational diagnosis, priority selection, and short-horizon action planning in small industries. Using a Design Science Research approach, the study develops a documented artifact corpus comprising a master instruction contract, 15 curated knowledge modules, structured output rules, a 60-scenario evaluation dataset, and automated runner and judge scripts. The artifact formalizes bounded operational reasoning through minimum-field gating, evidence separation, Knowledge-Resource Advantage synthesis, feasibility-weighted priority ranking, stopping rules, and conjunctive batch evaluation thresholds. A technical verification layer was conducted on a three-scenario local-smoke subset using smollm2:135m, qwen2.5:0.5b, and qwen2.5:1.5b with a heuristic judge. Results show that the evaluation scaffold is executable and that model capacity affects bounded instruction following, although hard-guardrail and priority-compliance failures indicate that the results should be interpreted as technical verification rather than final validation. The study contributes an audit-ready bounded Custom GPT architecture, a formal decision model, and a reproducible evaluation protocol for future comparative assessment of bounded and unbounded GenAI decision-support artifacts.]]>
Copyrights © 2025
