Responsibilities include:

• Design and fine-tune multimodal deep learning models, analyze video and text data, leverage annotated dataset for identifying harmful content in automated content moderation for TikTok and Reddit
• Manage analysis of over 35 million tweets to forecast community reactions to major sociopolitical events, enhancing fine-tuned LLaMA-3 models for charting change in public sentiment
• Analyze public opinion shifts across social media platforms and identify significant sentiment changes like a 22% increase in vaccine mandate support and a 25% rise in opposition to Roe vs. Wade from Twitter data
• Deploy fine-tuned GPT-based models on HuggingFace Hub and as Docker container, enabling a 30% reduction in model access time and seamless usage for collaborating teams at Northeastern University

Responsibilities include:

• Orchestrated an object detection pipeline on inhouse GPU infrastructure for analyzing product price trends in client’s competition
• Employed Text-to-Video Generative AI models to create promotional material for TCS New York City Marathon which was viewed by more than 2 million worldwide
• Mentored team of 5 undergraduate and doctorate interns, providing comprehensive support and offering research guidance, training and deploying machine learning models with on-premise GPU architecture
• Developed custom algorithm for title/heading extraction from scanned business documents with 80% accuracy and 25% lower inference time on edge devices using 16-bit quantization

Responsibilities include:

• Innovated and implemented novel, customizable functions leveraging Python, Matplotlib, and Seaborn for visualizing mean of rainfall variables across the Indian subcontinent over a 30-year period, accounting for leap year conditions
• Spearheaded experiments involving spatial and temporal normalization in super-resolution models
• Pioneered research efforts for identification of correlation between diverse climate variables and rainfall

Responsibilities include:

• Developed Flutter mobile app to calculate user attention with eye-line and directional data analysis using Facial landmarks.
• Achieved 25% reduction in system resource consumption using Provider state management and deferred loading , enhancing user experience and device longevity.