• Developed an end-to-end pipeline using gpt-3.5-turbo for automating feedback on system crash tickets by generating actionable summaries. • Achieved an 86% reduction in manual effort by providing comprehensive root cause analyses. • Attained 64.28% recall and 81.81% precision in classifying tickets as Valid Tickets or Noise Tickets. • Optimized the prompt and system processes, effectively reducing costs by minimizing prompt size and eliminating redundant LLM calls.

• Built a content-based Job Recommender System which recommends jobs to students based on their skills and professional experience. • Implemented the recommender using nltk and spacy ‘en core web sm’ model. • Developed Flask APIs which facilitated seamless storage of preprocessed descriptions within a MongoDB database and deployed the application on AWS ec2 instance

• Worked on a construction progress monitoring system using Deep Learning models.• Created a pipeline to process building component point clouds, generating vectors for classification using self-supervised models called Contrastnet and Clusternet and utilized SVMs for classification. • Used Stanford 3D Indoor Scene Dataset (S3DIS) for training and testing. • Improved model accuracy from 56% to 81% by adding heuristic-based hand-crafted features such as linearity, planarity, and verticality.