Objective • Built an interpretable profitability framework to identify the top 20% most profitable cardmembers from a 500,000- customer portfolio using 23 financial and behavioral features to optimize issuer-level Profitability optimization. • Achieved 0.905 leaderboard accuracy, improving over baseline by 0.42 point through reward-cost weight recalibration.

Objective • To develop a Deepfake detection framework capable of identifying manipulated videos under real-world distortions. • To improve model transparency by explaining spatial and temporal cues influencing Deepfake classification decisions. • Reduced attention complexity from O(T 2H2W 2) to O(T 2 +H2W 2), achieving state-of-art accuracy of 98.57% on FF++