Integrating Document AI and Vision-Language Models for Automated Electronic Manifest Generation

Generating accurate electronic manifests for international trade is crucial but hampered by error-prone manual processes. Artificial intelligence (AI) offers automation potential; however, integrating heterogeneous AI capabilities–specifically document AI and vision-language models (VLMs)–presents significant software engineering challenges often unaddressed by generic solutions. This paper contributes SmartManifest, a targeted software architecture framework designed to orchestrate these diverse AI components for automated electronic manifest generation. Employing a modular, layered, API-centric design informed by AI integration principles and comparative model analysis (e.g., LayoutLM, Donut, BLIP), the framework provides a replicable blueprint for this complex integration task. This paper details the architecture and reports on initial component validation using a prototyping methodology. This validation demonstrated technical feasibility (e.g., processing thousands of emails, generating relevant descriptions via BLIP) but also highlighted limitations (extraction fragility, lack of quantitative end-to-end metrics), defining specific requirements for future empirical work. The primary contribution is the architectural blueprint itself, addressing the specific software engineering challenges of integrating document AI and VLMs in this domain. This work establishes a methodologically grounded foundation for developing robust AI-integrated manifest systems and enables future comprehensive empirical evaluation.