Professional Summary
Undergraduate and early-stage ML Researcher focused on frugal AI for low-resource environments. Key works include developing an industry-certified frugal AI framework, being the lead author in a Nature Portfolio publication, and being selected as an instructor for IAIO 2026.
Education
BRAC University, Dhaka, Bangladesh
Bachelor of Science in Computer Science | May 2022 – Dec 2026 (Expected)
- Relevant Coursework: Statistics & Probability (4.0), Python Programming (3.7), Data Science (3.7)
Work Experience
Chowa Giken Corporation, Japan (Remote/OJT)
AI Engineering Trainee | Sept 2025 – Jan 2026
- Selected for a government-funded Japanese OJT AI Training Program (2% acceptance rate).
- Developed a novel 3D vision-to-numerical AI conversion algorithm (G3D-GVE) to estimate volume from 3D point cloud data without heavy image segmentation.
- Impact: +53.85% accuracy improvement, 98.33% reduction in training time (CPU inference: 2s), ~74.19% cost reduction vs standard 3D vision models.
- Awarded Best Group for architectural innovation and program performance.
- Patent under preparation, Main Inventor (Assignee: Chowa Giken Corporation), 2026.
Chowa Giken Corporation, Japan (Remote)
Incoming Machine Learning Engineer Intern | June 2026 – December 2026- Selected as an intern from a cohort of 20 trainees in the Nippon AI Dojo program.
- Role involves Applied Scientist responsibilities, including developing machine learning models, conducting experimental evaluations, optimizing model performance, and collaborating on real-world AI solutions.
Awards & Leadership
- Instructor, International AI Olympiad (IAIO) 2026 Camp: Invited by Dr. Golam Rabiul Alam (Top 2% Researcher, Stanford/Elsevier) to train national participants alongside faculty from Dhaka University and BRAC University.
- Founder, Epoch One (BRACU Independent AI Community): Launched a pilot AI community to enhance long-term AI talent in Bangladesh.
- Top 1% Learner, DataCamp (2024): Completed 60+ courses (ML Scientist, Statistician tracks).
- Best Contributor Award, Bengali Wikipedia (2021): Authored 50+ articles.
Research Publications
Journal
Advancing Cardiovascular Disease Diagnosis with Responsible AI Framework
Accepted: Scientific Reports (Nature Portfolio)
- Developed a tabular neural network ecosystem achieving 89% accuracy.
- Counterfactual explanations to suggest actionable changes for reducing cardiovascular risk.
- Aligned with FDA Good ML Practices and EU Trustworthy AI guidelines.
- Distinction: Lead author of a peer-reviewed publication in \textit{Scientific Reports} (Nature Portfolio) as an undergraduate.
- Read Paper |
Conference
BE-KNN: An Efficient Extension of k-Nearest Neighbors via Stable and Deterministic Ensemble Learning
Accepted: IEEE QPAIN 2026
- Supervised Epoch One AI Research Unit.
- BE-KNN: deterministic batch-based KNN achieving 12× faster runtime than stochastic ensembles, F1-score 0.94 vs 0.91 standard KNN.
Distinguishing Mainshocks from Foreshocks Using Spatiotemporal Labeling and Interpretable Ensemble Learning
Accepted: IEEE QPAIN 2026
- Developed spatiotemporal earthquake mainshock detection framework.
- Gradient boosting + Bayesian optimization, recall 0.97, AUC 0.965, SHAP-based interpretability confirmed magnitude differential as dominant predictor.
Preprints
JusticeNetBD — SSRN Preprint
- Retrieval accuracy: Recall@2 = 0.90, MRR = 0.90; ROUGE-L = 0.463, BERTScore F1 = 0.896.
- 23–25% improvement vs general-purpose LLMs; 10× faster response (1–2s).
- Read Paper | Live Demo
CausalDRIFT — MedRxiv Preprint
- Causal feature selection algorithm using FWL theorem & DML for healthcare ML datasets.
- Read Paper
Current Major Works
- Book Authoring: Co-authoring Practical AI for Bangladesh with Dr. Md Manjurul Ahsan (UN Board of Directors, Top 2% Researcher). Proposal submitted to Springer.
- Undergraduate Thesis (SABA-Net): CPU-efficient, zero-cost mathematical image classification model. 85% faster training on CPU vs GPU, competitive accuracy (0.90 vs 0.99). Code | Working Paper
Teaching Materials
- Authored Python guidebook widely used at BRAC University. Read Book
- Produced AI, Data Science, Statistics & Biostatistics video lectures. Watch Videos | Biostatistics Materials | Statistics Materials
Open Source Projects
CausalDRIFT (causalsoap) [PyPI] [GitHub]
- Python library for causal feature selection implementing CausalDRIFT algorithms.
Linear Coefficient Shift (linear-drift-detector) [PyPI] [GitHub]
- MLOps Python library for concept drift detection using OLS to monitor structural feature-target changes.
Technical Skills
- Languages: Python, R, SQL, Julia
- Machine Learning: Scikit-learn, TensorFlow/Keras, Explainable AI (SHAP), Counterfactuals (DiCE), Mathematical Optimization (PuLP)
- Data Science: Experimental Designing, Statistical Analysis, Applied Statistics
- GenAI & NLP: RAG, Prompt Engineering
- MLOps: Drift Detection
- Development: Streamlit, Render, Git
Selected Projects
Unbiased Journal Comparator Chatbot [Live Demo]
- AI system comparing research papers via Llama3 and DeepSeek R1, ignoring author reputation bias.
Advanced AQI Forecasting System [Live Demo]
- 6-hour forward-looking air quality predictor with explainable AI (SHAP) capabilities.
Lung Cancer Prognosis System (Bengali UI) [Live Demo]
- Accessibility-focused ML prediction system designed for Bengali-speaking users.
Warning System for Harassment [Live Demo]
- Prototype safety risk classifier using synthetic data generation and SHAP for model interpretation.
Contact
🌐 GitHub · LinkedIn · Google Scholar