At Ela•tEngineered, I built a multi-stage visual verification pipeline for return cases, combining CLIP/FashionCLIP similarity with Gemini 2.5 Flash review. After identifying that background composition could mislead CLIP, I added background suppression, colour-histogram filtering, and category-specific embedders, improving match precision by 31%.
At Log•kDeveloped, I developed Python chatbot logic with SQL-based retrieval across 30+ query types. I refined queries and workflows to reduce response latency by 20% while improving accuracy and reliability.
For CodeQuest, I built a full-stack competitive coding platform with real-time contests and an AI coding coach. The work was published and presented at ICCET 2026 as a peer-reviewed, Scopus-indexed paper; I also built a deepfake detection system that achieved 87.3% accuracy on FaceForensics++.
