Research Projects & Thesis (Masters)
Perishable Inventory Management Dashboard (Power BI) (Retail)
- Objective:
Enable small‑scale retailers to move from manual spreadsheets to an interactive, data‑driven approach for tracking, forecasting, and replenishing perishable stock—minimizing waste and stock‑outs.
- Solution:
Developed a Power BI analytics dashboard that:
Ingests sales, spoilage and supplier data via Power Query
Models it into a star schema (fact “Sales” table + “Products,” “Suppliers,” “Calendar” dimensions)
Defines dynamic DAX measures for spoilage rates, sales velocity, and turnover gaps
- Key Features:
Real‑Time Expiry & Stock Alerts: Automatic low‑stock and impending‑expiry warnings, delivered via Power BI subscriptions or mobile alerts.
Spoilage Analytics: Interactive treemaps and bar charts showing spoilage cost by category and product, plus “spoilage‑to‑sales” efficiency matrices.
Time‑Series Demand Forecasting: Built‑in line charts and heatmaps slicing by day‑of‑week and season to expose peak demand periods.
Supplier Performance Comparison: Clustered visuals comparing sales volume vs. spoilage rate for national vs. local suppliers.
Promotion ROI Dashboard: Scatter plots correlating discount levels with sales lift and post‑promo spoilage “echo.”
Turnover Gap Heatmaps: Matrix visuals of Days‑in‑Stock distributions to flag slow‑moving SKUs.
- Technologies Used:
Power BI (Power Query & DAX), CSV/Excel data sources (POS exports), Azure SQL (or local SQL Server), Power BI Service for sharing and mobile notifications.
Real‑Time E‑Commerce Personalization via Hybrid Recommendation Engine (Retail)
- Objective:
Overcome the limitations of static, batch‑mode recommenders by delivering highly relevant, context‑aware product suggestions in real time—boosting engagement and conversion on e‑commerce platforms.
- Solution:
Designed and prototyped a hybrid recommendation architecture that seamlessly ingests streaming user interactions, updates models on the fly, and serves low‑latency predictions through a scalable pipeline.
- Key Features:
Streaming Data Ingestion: Captures clickstreams, searches, add‑to‑cart events, and purchase logs via Apache Kafka for immediate processing.
Explainable Recommendations: Generates feature‑level attributions (via attention scores) so stakeholders can interpret “why” an item was suggested.
- Technologies Used:
Data & Streaming: Apache Kafka, Apache Flink (or Spark Streaming), Redis/Feast feature store
Modeling: Python, StreamLit
Deployment: Docker, AWS Lambda for realtime scoring
Evaluation & Monitoring: Grafana dashboard