Data Analyst · Information Systems · Applied Machine Learning
Michigan, United States
I work at the join between data analysis and applied machine learning.
The analysis side is the day job: cleaning and reconciling data nobody trusts, building the reporting that replaces a spreadsheet somebody rebuilds by hand every month, and making sure the numbers can be defended when someone questions them. MS in Information Systems from Central Michigan University, concentrating in Business Data Analytics and Project Management.
The research side is five co-authored papers on classification — sentiment analysis of product reviews, and deep learning applied to clinical images.
The thread running through both is the same, and it is not modelling. It is knowing which number to report. Accuracy on a 50:1 imbalanced dataset, confidence without lift, an AUC with no threshold attached — each of these looks like an answer and is not one.
| Repository | What it does |
|---|---|
| Retail Analytics Pipeline | Replaces a hand-built monthly Excel pack with extract → validate → transform → aggregate → publish. Fails before it publishes; every run leaves a provenance manifest. |
| SQL Analytics Toolkit | Cohort retention, RFM, funnel, Pareto and gap detection in portable SQL. Includes the two traps that produce plausible wrong answers: retention denominators and RFM recency direction. |
| DataQA | Declarative data quality checks. Written after a report was wrong for six weeks because an export started sending empty strings instead of nulls — which no null check finds. |
| Customer Churn Analytics | Ends where most churn projects stop too early: calibration, and the threshold as an economic decision rather than 0.5. |
| Market Basket Analysis | Apriori and FP-Growth implemented directly and cross-checked against each other. On why confidence is misleading and lift is not. |
| Project Risk Register | Java. Quantitative risk register, earned value, and PERT Monte Carlo scheduling — because 1–5 ordinal risk scores cannot answer "how much contingency do we need". |
| Repository | Venue |
|---|---|
| Amazon Reviews Sentiment | ICECIT 2021 · IEEE |
| SeAttnFusionNet | Discover Analytics, 4(1), 2026 |
| Hybrid AI Brain Tumor | Eastasouth J. Information System & CS, 2(02), 2024 |
| Skin Lesion 7-Class | Eastasouth J. Information System & CS, 1(02) |
Full record on Google Scholar
Several of these repositories are built around a metric being wrong rather than a model being better.
Accuracy on a seven-class dermoscopy set where one class is 58× another. Rule confidence when the consequent appears in 80% of baskets anyway. An AUC quoted with no threshold, so nobody can act on it. Percent-complete on a project that is behind — without earned value you cannot tell whether it is slow or expensive, and those need opposite responses.
Each of those produces a number that looks like an answer, passes review, and sends people in the wrong direction. Catching them is most of what an analyst is actually for.
Analysis Python · pandas · NumPy · SQL · R · Excel
ML scikit-learn · PyTorch · SciPy
Visualisation Tableau · matplotlib · seaborn
Databases MySQL · PostgreSQL · SQLite · MS SQL Server
Languages Java · Python · SQL · R · C++
Practice Business analysis · Agile/Scrum · risk management · planning
MS, Information Systems — Central Michigan University, Michigan (2025) Business Data Analytics · Project Management · GPA 3.52
BSc, Computer Science & Engineering — BRAC University, Dhaka (2021)
Junior Data Analyst — Warden Tech Ltd, Dhaka (2022–2023) Tableau dashboards replacing manual reporting · SQL and Python data preparation · source and method documentation across 5+ teams
Open to data analyst and analytics engineering roles.