Intern in the Systems Department - AI and Data Science Section

## Location ESTEC, Noordwijk, Netherlands ## Our team and mission The Systems Department is responsible for Systems Engineering, its implementation within Projects and In Orbit Demonstration and the innovative solutions - in terms of technology and methodology - enhancing its effectiveness: Artificial Intelligence, Advanced Simulation, Augmented Reality, Concurrent and Collaborative Engineering, Cost Engineering, Digital Engineering, MBSE, Sustainability.

Standardisation is also within the responsibility of the Systems Department, ensuring that processes are formalised and shared cross the broad community of space stakeholders.

The Future Engineering Division (in support of the programme directorates) is responsible for developing and providing the right Engineering Tools, Methods and Standards: - leading the Agency's centre of excellence for Artificial Intelligence and processing technologies for space applications; - leading the centre of excellence for Advanced Simulation Methods, including scientific data processing and databases, software applications including human-computer interfaces, support for science operation centres and downstream applications; - supporting projects with the centre of competence in Sustainability Engineering for all areas related to Clean Space, Life Cycle Analysis, EcoDesign, Zero Debris, REACH and RoHS; - performing pre-Phase A mission and system design studies in support of programmes' preparation of future missions, providing cost assessments and developing next-generation methodologies including Concurrent Design Facility (CDF), MBSE, and associated ontologies; - supporting the evolution of ECSS standards supported by all TEC departments and in full coordination with programmes.

The purpose of the AI and Data Science Section is to implement the ESA AI Competence Centre and lead the Agency's centre of excellence for advanced technologies supporting the development and exploitation of future space missions, contributing to the main mission of the Technology, Engineering and Quality Directorate in supporting European projects and their success through excellent engineering and the provision of innovative and enabling technologies, processes, methods, tools and applications.

The AI and Data Science Section carries out technology research in the field of Artificial Intelligence, Data Science and related areas.

Candidates interested are encouraged to visit the ESA website: http://www.esa.int ## 1) Topic 1: Knowledge Graph-powered Agentic workflows for Space Applications This internship first explores methods for combining knowledge graphs with agentic workflows (e.g., graph-as-memory, Agentic Graph RAG, or graph-as-coordination-hub for multi-agent systems) and maps these methods against candidate use cases in the space domain.

Building on this exploration, you will then design, develop and evaluate a prototype implementing one selected use case, demonstrating how the chosen method addresses a concrete space-related challenge. ## 2) Topic 2: Synthesis of Pareto-Optimal Interpretations for Neural Network-Based Anomaly Detection Systems This internship explores the synthesis of Pareto-optimal interpretations for neural network-based anomaly detection systems intended for autonomous onboard satellite operations.

The objective is to investigate methods for generating simple, human-understandable representations (such as decision diagrams) of complex black-box models (such as neural networks), while balancing competing objectives including predictive fidelity, interpretability, and representation complexity.

Constructing transparent and trustworthy interpretations is a key requirement for machine learning components in mission-critical domains, where accountability and explainability are essential.

You will tailor state-of-the-art symbolic reasoning techniques to machine learning models relevant to ESA programmes, with a particular focus on neural network-based onboard anomaly detection. ## 3) Topic 3: LLM-Assisted Verified Code Generation The increasing complexity of safety-critical software for space systems calls for new development methodologies that combine automation with rigorous correctness guarantees.

This internship investigates the role of Large Language Models (LLMs) as assistants in the development of formally verified software components for critical systems. You will explore LLM-assisted formal specification, where natural-language system requirements are translated into precise formal properties.

Building on these specifications, the internship will study approaches for automated code generation and refinement, using formal verification techniques to validate correctness, safety, and reliability.

The objective is to move beyond treating LLMs as autonomous code generators and instead establish a workflow in which LLMs serve as efficient engineering assistants whose outputs are systematically checked and validated through formal methods. ## 4) Topic 4: Symbolic AI to Simulate Quantum Convolutional Neural Network for Satellite Data Classification Quantum machine learning (QML) is emerging as a promising technology in image classification and remote sensing tasks that naturally arise in Earth observation.

However, due to current hardware limitations, the quantum circuit embedded into the AI pipeline is limited in size and the gate basis that it can use. We can still assess the potential advantage of QML approaches using simulators for the quantum circuit. In this internship, you will integrate novel quantum-circuit simulators based on automated reasoning techniques into the QML pipeline to classify satellite images.

The automated reasoning approach is agnostic to the gate basis used by the circuit, allowing quantum circuits of different sizes and with different gates to be implemented and evaluated in this context.

The outcome of the project will help us prepare for the time when quantum computers are commonly available and provide you with a unique opportunity to apply cutting-edge machine learning and quantum computing technologies to real satellite data. ## 5) Topic 5: Agentic-LLM ROS2 Controller for Rover Autonomy Increased autonomy of rovers would allow for further scientific exploration.

This internship explores the potential role agentic Large Language Models could play in enabling this autonomy, as well as its potential limitations. In this internship, you will implement an agentic LLM model framework that can take as input a task list and subsequently issue commands to the rover (using the ROS/ROS2 library) to accomplish these tasks and test the resulting quality of this framework.

The internship objectives include integrating pathfinding and collision avoidance tools and models into this agent; deploying this agent both in a lunar simulation environment and on board an analogue rover for performance evaluation; and exploring which stopgaps to implement to minimise machine error. ## 6) Topic 6: Evaluation Methods for Agentic AI in Systems Engineering Agentic AI systems are increasingly capable of performing complex engineering tasks involving information retrieval, reasoning and the use of external tools.

However, systematically evaluating their correctness, consistency and robustness remains challenging, particularly for tasks where results must be supported by different engineering sources. In this internship, you will investigate evaluation methods for agentic AI applied to systems engineering tasks.

You will define representative tasks and evaluation criteria, and develop a reproducible framework for comparing different agent configurations, models and reasoning strategies. The evaluation will consider representative activities such as requirements analysis, traceability assessment and verification support.

The outcome of the project will be a reusable methodology for assessing the capabilities and limitations of agentic AI in engineering applications, with the aim of supporting en