Thesis Intern Financial Crime Technology

• Conduct research for your master's thesis on representation learning and deep learning architectures to develop behavioural fingerprints for financial economic crime detection.
• Compare different model architectures, optimise performance within resource constraints, and evaluate models against FEC-related use cases.
• Collaborate with Deloitte's data science team and the partner Dutch financial institution to identify new use cases, discover emerging crime typologies, and refine model applications.
• Present research outcomes, model findings, and recommendations to support the development of an internal foundation model for FEC prevention.
• Join the Financial Crime Technology team at Deloitte, a supportive and dynamic working environment where your ideas are valued.

Naturally, you are keen to discover and learn new things each day and you are confident. But most importantly, be the true you. The one and only you. With your personal strengths, view of the world and unique personality. You need the following qualifications for the position of Thesis Intern Financial Crime Technology.

• Currently pursuing a Master’s degree in Artificial Intelligence, Computer Science, Data Science, Applied Mathematics, econometrics or a related field.
• A strong background in programming, statistics, mathematics.
• Good communication skills, curious mindset and eagerness to learn in a client-facing, team environment.
• Interest in preventing financial crime and translating research into real-world impact.
• Quick learning ability, high level of motivation and full-time availability (hybrid, with 2-3 office days).

We also value

• Previous experience with transformer-based models and representation learning
• A strong interest in or previous experience with agentic AI usage

Money laundering, terrorist financing and other forms of financial economic crime (FEC) are trillion-dollar businesses that fund large-scale criminal networks, erode trust in the financial system and undermine the legitimate economy. Consequently, identifying and preventing FEC is a core responsibility of modern financial institutions in the Netherlands.

To effectively do so, they critically rely on the ability to compare and analyse the behavioural patterns of millions of bank accounts. However, an account’s transactional history is complex, high-dimensional, and noisy, making it exceptionally difficult to derive a simple, standardised, and meaningful description of its true behaviour. This complexity prevents investigators from efficiently exploring patterns in account behaviour when investigating sophisticated crime typologies.

Representation learning can be used to address this problem. In this paradigm, unsupervised and self-supervised deep learning models transform raw, high-dimensional transaction data into a concise, quantifiable "behavioural fingerprint". This compact summary can serve many purposes in Anti-Money Laundering (AML) efforts, enabling downstream analyses such as outlier detection to identify potentially interesting accounts, clustering for new typology discovery, and retrieval of accounts that are similar to known FEC cases.

This research gives you hands-on experience with real-world data challenges, and the opportunity to contribute to an impactful project. You will also receive mentorship from industry experts in data analytics, AI, and the financial crime domain.