Design and Implementation of an LLM-based Order Management Module for Transportation Management Systems

This thesis investigates the controlled integration of Large Language Models into enterprise logistics workflows through the design and implementation of an LLM-based Order Management Module embedded within a Transportation Management System. The study addresses the structural gap between unstructured email-based transport requests and formal system registration processes, where manual interpretation constitutes an operational bottleneck. The proposed solution introduces a language - model - driven interpretation layer that extracts structured transport attributes from email messages and generates validated draft records prior to official system registration. The architecture separates probabilistic text processing from core business logic through strict JSON schema enforcement, backend validation rules, synchronized frontend - backend data structures, and a human-in-the-loop confirmation mechanism. This design will govern that the outputs of models are validated prior to persistence to maintain the governance criteria and accountability of operation. The evaluation was conducted using progressively complex test scenarios across both remotely hosted and locally deployed models. The results indicate that high-capacity remote models achieved greater structural consistency and extraction accuracy, whereas locally deployed models demonstrated weaker adherence to prompt constraints. Nevertheless, in all scenarios, the module significantly reduced manual transcription effort by converting free-text inputs into structured, machine-readable proposals. The results show that language - model - based extraction can be successfully integrated into enterprise platforms with the help of layered validation, regulation of execution limits and joint human control. Rather than pursuing full automation, the proposed approach establishes a transparent AI-assisted workflow that enhances operational efficiency while maintaining expert control.