>_ ML Researcher · Data Scientist · PhD Applicant
Building statistically rigorous frameworks at the intersection of machine learning, causal inference, and computational social science.
I'm a Machine Learning Researcher and Data Scientist based in Dhaka, Bangladesh, with a B.Sc. in Computer Science & Engineering from RUET (Thesis Grade: A+). My work focuses on developing statistically valid frameworks — not just accurate models, but ones with rigorous assumption validation, uncertainty quantification, and fairness auditing.
I'm currently a Researcher at Royal Scientific Publications and applying for PhD programs in Computational Science. I've won the Best Research Paper Award at APMEE 2025 and have two more papers accepted for presentation at IEEE SPECTRA 2026.
Beyond research, I'm a published poet, former radio jockey, lyricist, and composer — which informs the clarity and narrative depth I bring to communicating complex ideas.
AI-Powered Statistical Analysis Platform
Full-stack automated web app that transforms raw CSV data into professional statistical reports. 10 statistical tests with BH-FDR correction, 15+ visualizations, AutoML training 6 models, PDF/Excel export — all through a no-code dark-themed interface.
Persistent Memory RAG Chatbot That Never Forgets
Production-grade RAG chatbot with Ebbinghaus memory decay curves, ChromaDB vector storage, and Phi-3-mini running locally. Memories persist across sessions, naturally fade over time, and get reinforced on retrieval — just like human memory.
Feature-Wise Adaptive Imputation
ML framework that automatically selects the optimal imputation method per feature by learning a selector that maps feature statistical properties to the best strategy. Consistently within 1% of the best baseline — zero manual tuning.
Fairness-Aware ML for Dhaka Metropolitan Area
6-class crime prediction on 3,000 incidents using spatiotemporal cross-validation, SMOTE-NC, SHAP feature selection, Platt scaling calibration (ECE=0.0432), and FairLearn fairness auditing.
Bootstrap-Validated SHAP Framework for 169 Nations
End-to-end pipeline: PCA → K-Means clustering → XGBoost classifier with novel Bootstrap SHAP uncertainty quantification (200 iter). Only 6–8 of 20 features are truly statistically significant. Validated against EPI 2020 (r=0.76–0.82).
DistilBERT Fine-tuning + Classical ML Comparison
Two-part NLP project: fine-tuned DistilBERT on 10K reviews (87.05% accuracy), and a classical ML pipeline on 50K reviews with Logistic Regression achieving 89.76% accuracy. Production-ready .pkl artifacts included.
6th Annual Paper Meet Electrical Engineering Division (APMEE 2025) · Oral Presentation
Symposium on Photonics, Emerging Computational Technologies, Research & AI-Data Science (SPECTRA 2026) · IEEE Photonics Society HSTU SBC · Track: AI & Data Science
Symposium on Photonics, Emerging Computational Technologies, Research & AI-Data Science (SPECTRA 2026) · IEEE Photonics Society HSTU SBC · Track: AI & Data Science
Manuscript Under Review · First Author
Manuscript Under Review · First Author
Two versions available — one tailored for academic/PhD applications, one for industry data science and analytics roles.
Open to PhD opportunities, research collaborations, and data science roles. Feel free to reach out.