What I learnt building a career at the intersection of compliance, AI, and research

As a global financial compliance & crime associate at JPMorgan Chase, I investigate high-risk financial crime and file the reports that feed directly into federal law enforcement intelligence. It's demanding, detail-heavy work, and most days, it's invisible by design. A good investigation prevents a headline rather than making one.

Alongside that career, I have spent the last several years publishing peer-reviewed research applying artificial intelligence (AI) and machine learning to financial crime detection, testing how far these methods can be pushed beyond what conventional rule-based compliance systems can see. Looking back, a few lessons stand out from that work, and I want to share them in case they're useful to anyone building a similar path.

Tap here to add The Daily Star as a trusted source

Let AI-driven thinking guide your investigative work – and vice versa

Anti-money laundering (AML) compliance is fundamentally a pattern-recognition problem, and traditional AML systems are built on fixed rules: if a transaction matches a known pattern, it gets flagged. The limitation is obvious once you've worked in the field – rule-based systems only catch what they were already told to look for. That gap is what drove much of my research.

One study I am particularly proud of built an unsupervised machine learning ensemble to detect new money-laundering typologies within transaction graphs – patterns a system trained only on known, labelled laundering behaviour would never catch, because they don't resemble anything it has seen before. That's close to the exact problem I evaluate in live investigations, where the most dangerous cases are often the ones that don't fit any existing template.

I followed that work with related studies applying machine learning to credit card fraud detection and corporate bankruptcy prediction, and later to equity market forecasting using a hybrid ensemble model. In each case, I tested multiple algorithms – logistic regression, random forest, gradient boosting, XGBoost, and LightGBM – against real datasets rather than betting on one method. After all, in financial crime detection, a wrong model isn't an academic error; it's a missed case or a false accusation.

My advice: don't treat your AML expertise and your AI research as separate skill sets. The strongest financial crime research comes from investigators who understand both the criminal behaviour and the modelling well enough to know where each one breaks down.

Publish even when nobody is telling you to

Nobody required me to formalise my AML detection work into a peer-reviewed paper on emerging laundering typologies or to test ensemble AI models for fraud and market forecasting outside work hours. I did it because I wanted proof that the machine learning approach I trusted in live investigations would hold up to scrutiny from researchers with no stake in my day job. If you're wondering whether your AI-driven compliance work is "publication-worthy", my honest answer is you won't know until you write it up and submit it. Some of my early submissions were rejected. That's part of the process, not a sign to stop.

Practically, here's what worked for me in the AML and financial crime space specifically: start with a real detection problem you already understand from your job – laundering, fraud, or default risk. Then, find a dataset that reflects it realistically and run a rigorous, comparative test of several machine learning approaches rather than just one.

Reviewers in this field respond to honest, methodical model comparison far more than to a single flashy accuracy score.

Protect what you build

One thread of my AML and fraud-detection research eventually became a patent filing for a network security verification device, built on the same anomaly-detection logic I had already validated in my money-laundering typology research. The patent came from asking, after publishing, "Could this AI-driven detection logic be built into something deployable beyond a research paper?"

If you're doing applied AI research in fraud, risk, or security, ask that question of your own work. Not every paper needs to become a patent, but the strongest detection methods usually can.

Stay disciplined about your core field

It would have been easy to drift once I had a stable AML job, but I made a point of keeping my AI and machine learning research anchored to financial crime and financial risk – laundering typology detection, credit card fraud, bankruptcy prediction, and market forecasting. That focus meant each new paper built on the last, rather than starting from zero. My suggestion, especially for those early in a financial or compliance career: build depth with AI applied to your core field first. A stack of related, rigorous studies is worth more than a scattered handful of unrelated ones.

Build a body of work, not just a resume line

If I could go back and give my younger self one piece of advice, it would be to think in terms of a portfolio, not individual achievements.

A single AI-driven fraud-detection paper is a data point. A sequence of studies, like laundering typology detection, credit card fraud, bankruptcy prediction, etc., is a body of work. It compounds, and it becomes far more persuasive than any one credential on its own, whether you're applying for a job, a visa, or simply trying to convince yourself you're on the right track.

For anyone reading this and wondering where to start: pick one real detection or risk problem you already understand well from your job or your studies. Test more than one machine learning approach against it before you trust your conclusion, and write it up properly. Then ask what the next version of that idea could be – a related paper, a patent, an adjacent detection problem. Repeat that a few times, and a career takes shape that looks, in hindsight, a lot more deliberate than it ever felt while you were living it.

The author is a global financial compliance & crime associate at JPMorgan Chase. He holds an MS in Business Analytics from the University of North Texas, a BBA from American International University-Bangladesh, and completed a scholarly exchange as an Erasmus Scholar at Mykolas Romeris University in Lithuania.