AI Agent Development with LLMs: - Designed and developed standard framework for autonomous AI agents powered by LLMs (Large Language Models). Notably Agent Arya, which is engineered for high performance in Kaggle competitions, can execute end-to-end tasks, highlighting proficiency in diverse AI challenges.Top 15 Finish in Algonauts 2023 NSD Competition:- Secured a top 15 global ranking in the Algonauts 2023 challenge. Designed a transformer-based encoder-decoder framework to predict brain responses to diverse visual scenes, collaborating with Fractal's CelebralAI team and leveraging an extensive dataset from NSD.Ecommerce Image Generation Enhancement:- Contributed to advancing ecommerce solutions by enhancing image generation capabilities. Implemented cutting-edge methods to improve visual content creation, fostering a more dynamic and engaging online shopping experience.

Document Digitization- Led initiatives in document digitization, specializing in advanced table extraction leveraging Azure Form Recognizer.- Customized solutions for diverse formats. Introduced chart-to-table capabilities.- Implemented an intelligent algorithm for accurate page selection based on keywords. Developed a user-friendly web app for seamless demonstrations.AI-based Automation and Vendor Outreach- Spearheaded vendor outreach with a $50,000 budget, directly consulting with 5+ US-based vendors for data procurement.- Developed a product framework and data guidelines for a Computer Vision-based Instance Segmentation model designed Property Insurance sector- Optimized AI model deployment efficiency through substantial reduction in Docker Image sizeEcommerce Data Analysis- Automated data acquisition from IRI Unify using Python, overseeing 90+ IRI reports. Leveraged Snowflake to proficiently execute SQL queries for seamless data sourcing, pushing, and validation.- Implemented data harmonization using Concordia for multiple data sources, handling, exploring, and validating 10+ million rows of ecommerce data.

• Worked on backend and frontend development of web analytics dashboard for management of data pipeline supplied by IoT devices using Django• Reduced the data loading time by 25% using an optimized query to extract data from google cloud storage for web and application analytics dashboard• Improved the execution time 1.5 times by implementing multithreading to execute multiple queries and integrated it into custom-built REST APIs• Removed technical debt by 15%, refactored crucial codes, and implemented additional search filters improving lead management system dashboard

• Supervised a team of 4 to establish a pipelined process for automatic license plate recognition in videos captured by regular mobile phone cameras• Achieved a recall score of 64.24% for vehicle detection and 93.3% for license plate recognition using YoloV3 with a limited GPU resource of 2GB• Formulated vehicle tracking and tested in about 1500 video files obtained from AWS cloud storage integrated with the mobile application through APIs

• Worked on the problem of Visual Classification and Localization of Fashion-Centric images using concepts from Deep Learning and Computer Vision.• Developed and learned algorithms of Image Segmentation, Local Image Descriptors, and Clustering on image data using Python programming language.• Obtained an accuracy of 98% for Hand-Written Digit Recognition experimented on MNIST dataset by training a LeNet model using the Caffe framework• Attained an accuracy of 49.9% for apparel localization tested on DeepFashion dataset by implementing Faster RCNN using a variant of VGG network

• Elected to hall council by 750+ boarders, strategized and organized regular practice sessions, team meetings, and workshops for General Championship and Open IIT Events• Coordinated 5 teams with 12 captains and 90+ participants, ensured resource management during events, and bagged 1 gold, 1 silver, and 1 bronze medal

• Boosted mass reachability by designing creative portfolios, advertisements, and logos for both online and offline publicity and promotional purposes• Ensured smooth crowd management in arenas by quick decision making and consistent teamwork during the Global Entrepreneurship Summit 2015