Maahil Kureshi

I'm building a leakage-safe, time-aware content-scoring pipeline at FlyRank on a 79M-row search-performance warehouse, using DuckDB for out-of-core feature engineering and Scikit-Learn for classification.

For my Lenovo-collaborated MSc thesis at Middlesex University Dubai, I developed Time-Decay ALS for fashion recommendations. It improved MAP@12 by 86.59% over standard ALS on 31M+ H&M transactions.

I also implemented Neural Collaborative Filtering from scratch in PyTorch and benchmarked KNN, ALS, NCF, and TD-ALS across ranking, relevance, and diversity metrics.

My work spans Python, SQL, R, cloud platforms, dashboards, and applied analytics, including an Airbnb prediction model that achieved 84% held-out accuracy and Power BI reporting that reduced manual reporting time.