Nishant Agrawal

• Led development of AI/ML solutions for product retrieval, ranking, cloth detection, attribution, optimizing search & recommendation systems. • Managed end-to-end AI lifecycle, ensuring high accuracy & performance. • Integrated computer vision for real-time product categorization & LLMs for enhanced contextual recommendations. • Mentored engineering team, aligning efforts with company vision & technical goals. • Drove MLOps practices, optimizing resource utilization & streamlining deployment processes.

• Developed new features on blueprint services using Rastor and vector-based data, incorporating visual and text search. • Enhanced smooth edges in semantic segmentation for large aerial images. • Standardized development processes for improved workflow efficiency, incorporating Jira. • Implemented MLOps practices for optimal resource utilization.

• Enhanced ADAS performance by incorporating features like seatbelt, mask, and camera obstruction alert. • Optimized driver distraction and drowsiness detection performance. • Developed Facial Recognition System pipeline with automated training/inference, enhancing security and electronic driver logging. • Handled entire ML application lifecycle from research, model development, deployment, to performance monitoring and data management.

• Enhanced training speed by exploring different augmentation strategies, preprocessing methods, and regularization techniques. • Conducted a comparative analysis of these methods from existing literature. • Developed image retrieval via similarity index using self-supervision for embedding estimation. • Implemented color-based similarity in fashion domain data.

• Developed efficient technique for generating error maps via deep learning, enabling change detection in images. • Instituted a convolutional hybrid for autoencoder in encoder block, improving image encoding learning. • Applied fuzzy c-mean clustering on error map to produce binary difference map. • Models were coded in Python, leveraging keras API, and tested on various image datasets. • Model improvements detected changes in small pixel areas and vibrant shadows, often overlooked in prior models.