• Developed ’Escalation’ feature that improved Compliance module, reducing review cycle time by 35% via automated escalation

• Worked on time-based partitioning and indexing of NoSQL database, which improved retrieval efficiency by 4% at petabyte scale

• Reduced misflagging rate by 18% using finetuned BERT with class weighted loss and hard negative mining on near boundary samples

• Worken on post-ranking optimization for IoT recommendations using Upper Confidence Bound to balance exploration and exploitation

• Evaluated ranking quality using relevance and diversity, improved long-tail item exposure by 11%, keeping NDCG@50 within 5% of baseline