About Me
I am an ICT Assistant at the International Training Network (ITN) Centre, BUET, where I own the institute's digital platforms end to end — the official website and sanitation knowledge repository, two full online courses for government officials, and the data pipelines behind our district-level survey work.
Day to day that means developing and maintaining ITN-BUET's WordPress site with the Divi builder, handling site structure, content publishing workflows and social media channels. I built two end-to-end Moodle courses from scratch, authoring interactive H5P learning modules and configuring course structure, user enrollment and learner progress tracking. I also facilitate capacity-building training for municipalities on IMIS, CWIS, FSM and WASH — providing technical support and troubleshooting hardware, software and presentation systems — and coordinate a district-level government survey programme, processing and visualising the datasets in Python and KoboToolbox to produce analysis reports.
I came to this work from research. My Computer Science thesis at BRAC University — Context-Aware Zero-Shot Anomaly Detection in Surveillance — designed a dual-stream framework combining TimeSformer-based spatiotemporal transformers, DPC predictive modelling and CLIP vision–language alignment to detect unseen abnormal events without any labelled anomaly data. It reached 84.5% ROC-AUC and 72.3% PR-AUC on UCF-Crime, outperforming every other pure zero-shot baseline, at 0.45s detection latency.
Alongside it I have built applied ML pipelines end to end: a spectrophotometric framework evaluating 50 commercial blue-light blocking lenses across the 380–780 nm spectrum (KNN, R² = 0.914), and a home-loan eligibility classifier that handles severe class imbalance with SMOTE (Random Forest, 89.29% accuracy). Comfortable owning systems end to end, from WordPress and Moodle deployments to Python data pipelines, I am now moving toward a software engineering role building reliable software systems.