Intelligent Shipping Orchestration Platform

As shipment volumes grew, the client’s logistics network became too complex for static, rule-based automation. Exceptions such as customs delays, regulatory changes, and carrier performance variations required a more adaptive system. To solve this challenge, Leobit began introducing AI into the platform about a year and a half ago.

The first step was the deployment of a retrieval-augmented generation (RAG) system. Employees often had problems navigating customs regulations and international shipping requirements, which frequently led to costly mistakes. With the new system, they could ask natural language questions such as, “What documents are required to ship heavy copper vases to Germany?”

The solution retrieved relevant regulations from a curated knowledge base and used GPT-based models, orchestrated through LangChain, to generate accurate and context-specific answers. A vector database supported semantic search to ensure that responses were always grounded in reliable sources.

Leobit trained document classification models to automatically recognize customs declarations, VAT forms, and invoices, which cut manual review time by nearly 64%. Leobit also built a price calculator that combined shipment dimensions, weight, and declared value with carrier data to estimate costs and delivery times. Additionally, Leobit implemented an optimization layer that recommended the most efficient balance between cost and speed.

Thanks to the carrier integrations already established during the early years of the partnership, Leobit upgraded the platform to support fully automated carrier selection. The AI engine not only compared prices and estimated delivery times, but also evaluated historical carrier performance, service reliability, and customer preferences. By introducing this adaptive decision-making process, Leobit enabled the client to move far beyond the limitations of traditional rule-based automation.

The most transformative stage of the project involved the introduction of AI agents. The first agent we developed focused on logistics. It was designed to dynamically plan and execute workflows. For example, when a shipment encountered a customs delay, the agent could automatically prepare updated documents, notify the customer, and propose an alternative delivery route. This was made possible by combining a reasoning LLM with modular API connectors for ERP systems, carrier platforms, and communication tools.

Over time, this solution evolved into a multi-agent system. Instead of relying on a single agent to handle every task, Leobit created specialized agents for distinct domains. Our experts designed a logistics agent to manage routing and documentation, built a customer service agent to deliver real-time shipment updates and alternative delivery options, and developed a financial agent to automate billing and VAT processes. Leobit connected these agents through a shared orchestration layer, which enabled them to collaborate seamlessly, avoid conflicts, and deliver a unified operational experience.

The client’s platform has undergone a complete transformation in its architecture. Initially, it was a straightforward rule-based system built around dashboards, APIs, and document automation services. Today, it has evolved into a hybrid environment that combines deterministic automation with adaptive, agentic AI.

At the core, .NET and Python microservices drive processing logic, while React and Tailwind power the web interfaces. PostgreSQL remains the backbone for structured data, complemented by vector databases that support semantic retrieval for the RAG system. OpenAI GPT models, together with LangChain pipelines, enable reasoning, orchestration, and workflow adaptation. The platform is hosted on Azure, with Docker and Kubernetes ensuring scalability, resilience, and secure deployment.