Executive summary
The Port of Antwerp-Bruges, in collaboration with Superlinear, has embarked on a transformative journey to become a fully integrated Smart Port. Central to this transformation is the APICA Chatbot (the Advanced Port Intelligence & Coordination Assistant), a sophisticated AI-powered tool designed to streamline operations and improve efficiency. Here, we explore the technical specifics of how APICA Chat was conceptualized, developed, and implemented within the port’s infrastructure.
By creating APICA, we are not just optimizing port operations, we are fundamentally changing how our teams interact with technology. It’s about applying our principle of ‘placing people & customers at the center of everything we do’, whereby we aim to make every task frictionless and more intuitive,” said Pieter Van Bouwel, Team Lead Data Science of Port of Antwerp-Bruges.
Case study
The challenge
Employees in all organizations spend a significant amount of time searching for information and performing tedious tasks that could in theory be automated by LLM agents. However, the distribution and maintenance of these LLM applications across different platforms introduce their own set of issues.
Take, for example, the scenario where one LLM agent is dedicated to querying the HR knowledge base, another assists users with managing calendars and emails, and a third supports users with questions on operational data. If each of these agents operates on a separate platform, it creates a fragmented system that complicates user experience. Users struggle to locate the appropriate tool for their needs, while administrators face increased complexities and burdens in maintaining various systems.
This disjoint setup can lead to inefficiencies, as employees spend unnecessary time navigating multiple agents, and administrators struggle with the technical and logistical challenges of managing a wide array of agents. This limits the potential of LLM technology to streamline processes and improve organizational efficiency.
Conceptualization and scope
When generative AI started gaining traction, PoAB envisioned a future where scattered expertise could be unified into a single virtual colleague, leading to the creation of APICA chat. The APICA Agent was launched with a clear vision: to become the new digital colleague for every co-worker in the port. It assists with querying information and performing simple tasks, reducing manual workloads, speeding up decision-making, and enhancing data access efficiency.
A crucial aspect of this vision was the development of different LLM agents, each serving as a domain expert. These applications are specialized in querying specific data sources, such as HR data, operational information, or email and calendar management. By making all these applications available to an orchestrator agent, PoAB ensures that user queries are directed to the appropriate LLM domain expert agent.
Architectural Design
System integration
APICA Chat was integrated within Microsoft Teams, leveraging the platform’s widespread use across the port for a seamless adoption process. This integration allows employees to interact with APICA just like they would with any other colleague.
Building blocks
1. Orchestrator Agent
The orchestrator serves as the central point of access, responsible for interpreting user questions and routing them to the appropriate LLM domain expert agent. By understanding the intent behind each question, the orchestrator ensures that employees receive the response from the most capable LLM domain expert.
2. LLM Domain Expert Applications
These applications act as specialized domain experts. Currently, we have implemented the following three domain experts:
Nautical data expert: Functions as a business analyst, querying data catalogs and the data warehouse to answer nautical questions, such as how many seaships arrived yesterday afternoon through this particular lock.
ICT Service Assistant: Supports people with questions related to ICT concerns.
HR Assistant: Assists people with questions related to HR concerns.
Examples:
Nautical Data Expert
HR Assistant
In general, these domain expert agents can perform 2 type of tasks:
Information Retrieval and Preparation: The agent navigates through the data sources to acquire and extract the relevant information in a concise manner. For instance, if an employee needs details on vacation days, the agent quickly accesses HR data, interprets the necessary details, and delivers clear and timely information.
Task Execution and Automation (Agentic Operations): The capacity to perform tasks on behalf of the user, from executing SQL statements on a data warehouse to creating service desk tickets if unable to find an answer.
3. Python Library of Reusable Components
This library contains a collection of reusable components that facilitate the sharing of intelligence across different LLM agents. By ensuring that these components are reusable, the system promotes consistency and efficiency in deploying AI solutions, allowing for scalable and maintainable development.
How APICA grows her brain: future enhancements
The expansion of APICA’s capabilities is strategically managed through orchestration. APICA's “brain” includes several sub-domain LLM agents, each designed for specific tasks like information retrieval and agentic actions. APICA utilizes the most appropriate agent for each query.
Each agent is tailored to process a particular data source optimally. The modular setup allows for easy integration of new knowledge. Increasing the brain is thus simply a matter of adding a new LLM domain expert agent to the orchestrator.
To efficiently maintain and enhance this portfolio of LLM domain expert agents, we have created a template to quickly build new LLM apps, ensuring best practices within each LLM domain expert. In this way we can maintain uniformity across them. Improvements to one application become compatible across all, allowing developers and administrators to easily contribute to different apps. By using the same base, this approach streamlines development and maintenance, enhancing consistency and collaboration.
Conclusion
APICA Chat addresses the challenge of scattered expertise by serving as a unified virtual colleague, transforming how employees at the port access information and perform tasks. By integrating smoothly with Microsoft Teams, APICA reduces manual workloads, speeds up decision-making, and enhances data access. Its flexible design allows continuous growth and improvement, making it the ideal partner now and in the future.
