Profile
About
I'm an AI researcher and builder based in Dhaka, Bangladesh. My work sits at the intersection of applied deep learning and the practical question of whether a model does anything useful outside a notebook.
My undergraduate thesis compared two convolutional architectures, DenseNet121 and EfficientNet-B0, for detecting disease in corn leaves from images. It was my first sustained piece of research, and the most valuable part wasn't the modelling — it was learning how much of a result depends on decisions made before training starts: how the data is split, what is held out, and what the number on the screen actually describes.
I studied software engineering at Daffodil International University, graduating in June 2026. That background shapes how I approach research: I'd rather ship something reproducible than something impressive. It also means I can build the surrounding system — the interface, the packaging, the documentation — not just the experiment.
I'm at the beginning of a research career and I'd rather say that plainly than inflate it. What I have is a body of work I can explain end to end, and a clear direction: computer vision applied to problems where the data is messy and the stakes are real.
How I work
01
Report what the model actually did
A number is only useful if someone can re-run it. Datasets, splits and seeds get documented alongside the result, not after it.
02
Prefer the boring baseline
New architectures are easy to justify after the fact. I try to establish what a simpler model achieves first, then argue for complexity with evidence.
03
Build the thing, not just the notebook
A trained model that nobody can run is a claim, not a system. I care about the path from experiment to something usable.