● Engineered an end-to-end classification pipeline that processed over 13,000 customer support tickets, achieving a 100% automated classification rate and providing proactive insights into emerging issues. ● Developed a multi-stage data processing module using LLMs to automatically extract and normalize critical information, including site, MAC address, and issue description, from raw Salesforce ticket data. ● Implemented an advanced LLM-based analysis to parse support agent-customer conversations, automatically identifying root causes and agent actions, which helped streamline case resolution and improve the agent knowledge base. ● Leveraged BERTopic on a dataset of over 13,000 "complete issue" data points to uncover 56 distinct, actionable topics and 728 subtopics, enabling the model to adapt to dynamic support trends without manual retraining. ● Built and deployed a final classification model using Facebook BART to automatically route each ticket to the most relevant subtopic, ensuring every new ticket is assigned a precise classification from the unsupervised model's established categories.

● Developed machine learning models that ran 6000+ customer images per day in real-time to alert users regarding improper face uploads at the front end, increasing photo acceptance rate from 80% to 90%. ● Created a machine learning model using XGBoost to predict and identify churned users in advance having 93% recall, and led marketing plan campaigns. ● Devised CNN classifier for document recognition across 11 classes, achieving 97% precision and 92% accuracy. ● Engineered scalable ML models that handle 20000+ calls per day from the backend and frontend, automating the KYC verification process and streamlining the onboarding journey for users. ● Designed a K-Means clustering approach to segment 1.1M users according to investment habits in Mutual Funds. ● Initiated statistics-based verification of account statements for Income and Bank Proof, automating 85% of account statement pdf uploads.

● Designed ML models to extract Period and Names from Account Statements with 95% precision. ● Built ML model to perform similarity analysis using Random Forest Classifier with 98% precision. ● Deployed models as Rest API using FastAPI for scaling.

● Performed literature survey to identify gripper selection and pose estimation for robotic grippers in assembly lines. ● Created an AI-based system based on computer vision for job recognition to estimate object pose and coordinates with 100% accuracy to grasp on the test set. ● Published Real Time Deep Learning-Based Image Processing for Pose Estimation and Object Localization in Autonomous Robot Applications, The International Journal of Advanced Manufacturing Technology, Springer.