About

I’m a machine learning engineer at IconPro GmbH, where I work on industrial AI. Most of my work is about helping manufacturers reduce scrap: building predictive models of production processes and using reinforcement learning to find process settings that improve several quality targets at once. Alongside that, I work on explainable AI, to make the relationship between process parameters and product quality understandable to the engineers who run the line.

Working with models that make real decisions has made me curious about what is actually going on inside them. I want to understand how neural networks learn: which concepts they form, which mechanisms they use, and how we can check that they behave the way we think they do.

Background

I studied Computer Science and Engineering at BRAC University in Dhaka (B.Sc., 2013–2017). My bachelor’s thesis, a small first research project on predicting movie box-office success, led to my first two conference papers. I then spent two years as an executive engineer at Meghna Group of Industries, where I led IoT projects and built a fleet-monitoring platform for more than 2,000 trucks.

In 2019 I moved to Germany for my master’s, an M.Sc. in Data Science at RWTH Aachen University, and alongside my studies I worked part-time at IconPro. My master’s thesis developed a multi-objective reinforcement learning framework for optimizing production process parameters. The work continued after I graduated in 2023 and was published in Applied Sciences in 2025. You can read more about it on the research page.

Research interests

Mechanistic interpretability: reverse-engineering the circuits and features models use, with tools such as sparse autoencoders.
How networks learn concepts: how internal representations form during training and what shapes them.
Auditing and intervention: using model internals to find hidden or unsafe behavior, and in some cases to intervene on the specific mechanisms responsible.
Interpretability in practice: bringing these ideas to applied settings like industrial AI, where decisions need to be explainable to the people who rely on them.

Contact

Email me at [email protected], or find me here: