Data Projects

Online grocery pricing in Bangladesh moves constantly — platforms run rotating offers, and brand presence shifts week to week. This project treats that as a market-intelligence problem: scrape live pricing data, structure it, and surface the patterns a category manager would actually want to see.

Tools & purpose

• Scraped 8,085+ live product and pricing data from online grocery platforms (Chaldal & Shwapno) using Selenium.
• Cleaned and transformed data with pandas / numpy in Jupyter Notebook.
• Built an interactive Tableau dashboard visualizing pricing trends, offer & savings distribution, brand dominance, and market insights.
• Published the full project on GitHub.

Challenges in Data Processing

• Removed ~890 duplicate and null entries during data cleaning.
• ~350 multi-category products were standardized by keeping the most relevant category.
• Price and unit variations exist for the same product across platforms due to packaging differences.
• Brand extraction (~700 brands) required keyword-based parsing with manual correction.
• ~4,000 products with generic units were processed using derived fields (actual unit, savings, price metrics, etc.)
• Out-of-stock status from Chaldal could not be fully captured, which may slightly affect analysis.

What the dashboard shows

• Pricing trends across products and platforms over time.
• Offer and savings distribution — where discounts are concentrated.
• Brand dominance — which brands show up most often, and at what price tier.
• Overall market insights useful for spotting pricing patterns at a glance.

An end-to-end HR analytics project that tracks employee attrition, compensation, performance, diversity, and attendance using SQL for analysis and Power BI for visualization.

Tools & purpose

• PostgreSQL — structured data storage with 5 relational tables.
• SQL — 12 business-question-driven queries covering attrition, compensation, performance, diversity, and attendance.
• Power BI — interactive HR dashboard built on top of the query layer.

What's inside

What the dashboard shows

• Workforce size, active headcount & growth trend analysis.
• Employee attrition rate and retention pattern analysis.
• Department-wise attendance and attrition performance.
• Performance rating distribution and improvement trends.
• Salary analysis by department, gender & performance level.
• Employee tenure and retention behavior analysis.
• Gender diversity insights across departments.
• Top-paid active employee identification & workforce insights.

RetailNexus is an end-to-end data analytics project built on a retail sales dataset. It simulates a real-world business intelligence pipeline — starting from raw data cleaning in Python, structured storage in PostgreSQL, and finally an interactive multi-page dashboard in Power BI. The goal is to give business stakeholders a single unified hub to monitor sales performance, customer behavior, product trends, and financial health — all in one place, with interactive filters.

Tools & purpose

• Cleaned and structured raw retail sales data with pandas.
• Designed structured storage in PostgreSQL to mirror a production data pipeline.
• Built an interactive, multi-page dashboard in Power BI for sales performance analysis.

What's inside

Business questions answered

• Which cities and store branches generate the most revenue?
• Which customer segments (age group, gender, membership tier) are most valuable?
• What are the top-selling and most-returned products?
• How do sales channels (In-Store, Online, Mobile App) compare?
• Does offering a discount actually improve profit margin?
• Which months and days of the week see the highest sales?
• How is revenue trending over time vs. profit?

FinSight Banking Analytics is a full-stack data analytics project built on a simulated banking dataset of 2,150 customers. It covers the complete analytics pipeline — from raw data preprocessing to SQL-based analysis and an interactive Power BI dashboard — with AI tools integrated at each stage to accelerate development and improve accuracy. The project was built around a practical, real-world workflow where AI assistants are used as professional productivity tools, not replacements for analytical thinking — each one selected based on its fit for the specific environment and task.

Tools & purpose

• Cleaned and preprocessed raw customer data with Python (Pandas), assisted by GitHub Copilot for transformation logic and null handling.
• Built 13 analytical SQL queries in PostgreSQL covering segmentation, loans, balances, and regional trends.
• Re-implemented all 13 queries as DAX measures in Power BI Desktop, using Claude AI to resolve table conflicts, case-sensitivity issues, and SQL-to-DAX syntax differences.
• Designed an interactive Power BI dashboard with Power Query (M) transformations on top of the DAX layer.

AI-assisted workflow

• Stage 1 (Pandas): GitHub Copilot in VS Code accelerated data cleaning and preprocessing while data-quality checks stayed manual.
• Stage 2 (SQL): Copilot suggested query structure, aggregations, and CASE/GROUP BY logic; every query was reviewed before use.
• Stage 3 (DAX): Since Copilot isn't available in Power BI Desktop, Claude AI was used to convert and debug DAX measures — chosen specifically for the environment.
• Each AI tool was selected for task fit, with analytical decisions and final review handled independently.

What's inside

What the dashboard shows

• Customer segmentation by status, region, and account type.
• Loan penetration and ownership patterns.
• Monthly account-opening trends over time.
• Balance distribution across credit score bands.
• Regional financial performance comparison.

A statistical deep-dive into 10.48M mobile money transactions from the PaySim dataset, using descriptive statistics, outlier detection, hypothesis testing, and logistic regression to identify what separates fraudulent activity from legitimate transactions. The goal is to move past a simple classifier and surface the statistical signals — transaction size, type, and balance behavior — that actually distinguish fraud, and translate them into a monitoring recommendation.

Tools & purpose

• Analyzed 1,048,576 transactions from the PaySim synthetic financial dataset (Kaggle).
• Applied descriptive statistics and IQR outlier detection to characterize transaction behavior.
• Ran t-tests and ANOVA to test whether fraud transactions differ meaningfully in size and type.
• Built a logistic regression model to identify the strongest predictors of fraud.

What's inside

Business questions & methods

A collection of 5 frontend projects built to demonstrate responsive UI development, modern CSS layouts, and JavaScript-based interactivity.

• Travel Website Design :
  Responsive landing page with hero sections, destination grids, and modern layout design using HTML5 & CSS3.
  
  Live Preview 🔗
• E-Commerce Website :
  Online shopping interface featuring product cards, navigation structure, CSS Grid layouts, and responsive design.
  
  Live Preview 🔗
• Admin Dashboard :
  Analytics dashboard UI focused on grid positioning, data cards, tables, and dashboard-style components.
  
  Live Preview 🔗
• YouTube Interface Clone :
  Pixel-inspired frontend recreation with semantic HTML, sidebar structure, and responsive UI elements.
  
  Live Preview 🔗