MLOps with Microservices:A Case Study on the Maritime Domain

Renato Cordeiro Ferreira, Rowanne Trapmann, Willem-Jan van den Heuvel · 2025

Read MLOps with Microservices:A Case Study on the Maritime Domain by Renato Cordeiro Ferreira, Rowanne Trapmann, Willem-Jan van den Heuvel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-06202.txt.

Introduction

Renato Cordeiro Ferreira and two co-authors report a case study on building a machine learning system for shipping in this two thousand twenty-five paper. The system, Ocean Guard, spots odd patterns in maritime traffic that may signal trouble. It was built as a set of small services, called microservices. That split let several teams work in parallel. The paper tells how the team ran machine learning operations, or MLOps, under this design.

The key device was contracts between services. Separate contracts governed code, models, and data, so each team knew its duties. The authors share the hurdles they hit and the lessons they drew. Software and data teams building large learning systems are the audience. The case shows how contract-based design can keep such projects coherent.

Catalog record

CREATORRenato Cordeiro Ferreira, Rowanne Trapmann, Willem-Jan van den Heuvel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE27 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-06202

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MLOps with Microservices:A Case Study on the Maritime Domain Renato Cordeiro Ferreira, Rowanne Trapmann, Willem-Jan van den Heuvel Abstract This case study describes challenges and lessons learned on building Ocean Guard : a Machine Learning–Enabled System (MLES) for anomaly detection in the maritime domain. First, the paper presents the system’s specification, and architecture. Ocean Guard was designed with a microservices’ architecture to enable multiple teams to work on the project in parallel. Then, the paper discusses how the developers adapted contract-based design to MLOps for achieving that goal. As a MLES, Ocean Guard employs code, model, and data contracts to establish guidelines between its services. This case study hopes to inspire software engineers, machine learning engineers, and data scientists to leverage similar approaches for their systems. 1 Introduction Maritime transportation is the cornerstone of global trade and commerce, facilitating the movement of goods and commodities across vast distances. The maritime domain represents one of the most dynamic and complex operational environments in global logistics and transportation. It is characterized by continuous movement across international boundaries, varying regulatory frameworks, and vulnerability to unpredictable weather conditions. As such, maritime operations generate a large amount of heterogeneous data from multiple sources. The digital transformation of the maritime industry has accelerated in recent years, driven by economic pressures, environmental regulations, and safety concerns [ 3 ] . According to the International Maritime Organization (IMO), ships now function as floating data centers, equipped with hundreds of sensors. Improving maritime navigation and communication by facilitating the implementation of safe data sharing between vessels and between vessels and shore-based systems through automation [ 1 ] . This complex web of information flows across organizational boundaries between shipping companies, port authorities, maritime safety agencies, and commercial partners [ 8 ] . This rich data ecosystem presents opportunities and challenges, including the potential for predictive analytics, anomaly detection, and route optimization using Machine Learning (ML) [ 2 ] . However, extracting actionable insights requires systems capable of ingesting, processing, and analyzing different data streams at scale while maintaining operational reliability in an environment where connectivity and conditions rapidly change [ 17 ] . Machine Learning Operations (MLOps) refers to a collection of best practices, sets of concepts, and development culture that aims to achieve that goal [ 10 ] . Since 2015, the microservices architectural style have emerged as a promising approach to address challenges of modularity, scalability, and resilience by decomposing complex systems into smaller, independently deployable services with well-defined interfaces [ 14 , 16 ] . This approach can be particularly suited to the maritime domain’s nature, where specialized components can be designed to handle different maritime data sources and analytical needs. The novel field of Software Engineering for Artificial Intelligence (SE4AI) studies the adaptation of Software Engineering techniques for the creation of ML-Enabled Systems (MLES) [ 9 ] . Since 2022, different reference architectures have been proposed for them [ 11 , 6 , 10 ] . Fundamentally, they describe these systems as containing multiple components, which can be divided across different services. This case study describes the development of Ocean Guard : a ML-Enabled System (MLES) for anomaly detection in the maritime domain. Section 2 details the functional and non-functional requirements considered for the system. Section 3 gives an overview of the system architecture, designed with a microservices architecture for MLOps. Section 4 zooms in in the Ocean Guard API, designed with a hexagonal architecture with Domain Driven Design (DDD). Section 5 shows the different teams involved in the development of the tool. Section 6 describes how the developers adapted contract-based design to MLES for developing different components in parallel. Section 7 lists the main challenges during the development of the Ocean Guard tool, according to the experience of the authors. Section 8 then summarizes the key lessons learned, related to applying development techniques from microservices on ML-enabled system development. Finally, section 9 lists the main contributions of paper and future work. 2 System Specification This section describes the specification of the Ocean Guard tool, which was defined from conversations with stakeholders at the beginning of the MARIT-D project. The specification is divided into two parts: functional and non-functional requirements [ 15 ] . 2.1 Functional Requirements Functional Requirements , document the functionalities that directly affect the user of a system [ 15 ] . There are two actors related to the Ocean Guard tool. The following descriptions detail the main functionalities associated with them. An Investigator is a human user who uses the tool to obtain clues about anomalous behavior at sea. They must be able to: I1. see geolocations of marine objects in a map, I2. filter geolocations by area of interest, date and time, I3. discern different types of marine objects (vessels, structures, or unidentified), I4. retrieve geolocations gathered from different data sources, I5. check details (metadata) associated with a given marine object, I6. highlight the trajectory of a marine object, I7. see anomalies identified by the tool in a map, I8. filter anomalies by area of interest, date and time, and I9. inspect why an anomaly was considered so by the tool. The Anomaly Detector is a component of the tool responsible for detecting anomalous behavior of marine objects. It must be able to: A1. detect anomalies related to a marine object, A2. list anomalies by area of interest, date and time, and A3. explain why an anomaly is considered so. 2.2 Non-Functional Requirements Non-Functional Requirements , also known as Architectural Requirements , describe a set of nonstructural attributes that are important for the success of a system [ 15 ] . There are four main non-functional requirements proposed for the Ocean Guard tool. The following descriptions detail the main characteristics the system should have to attend its users. Compliance. The Ocean Guard tool should be compliant with the latest European laws, including the General Data Protection Regulation (GDPR), and the Artificial Intelligence Act (AI Act). Explainability. The Ocean Guard tool should be able to explain why it considers a marine object to be anomalous. For algorithm-based anomaly detection, this means clarifying which decisions were taken until the anomaly was identified. For AI-based anomaly detection, this means using interpretable models (such as tree-based models) or implementing black-box explainability techniques (such as SHAP or LIME). Resiliency. The Ocean Guard tool should be able to handle the expected amount of data. This is important because the system may process high-volume and high-velocity data sources, such as those providing tracking data. Compatibility. The Ocean Guard tool should be able to interoperate with other existing tools, extending the Investigator ’s current existing capabilities. As such, the tool should follow EU standards that enable quick integration with other ecosystems. 3 System Architecture This section presents the architecture of the Ocean Guard tool, designed to implement the requirements presented at section 2 . Figure 1 showcases the components planned for Ocean Guard , following the reference architecture representation proposed by Ferreira et al. [ 6 ] . Figure 1: System Architecture of the Ocean Guard tool This architecture follows the same notation introduced by Ferreira et al. [ 6 ] . Rectangles represent applications or services , which execute continuously. Stacked rectangles represent pipelines , which execute

Read the complete text