1. Perform qualitative and quantitative market research.2. Conduct focus groups, surveys and interviews.3. Conduct recitations and perform grading.

1. Improved operational efficiency at the $700M scientific user facility by developing and deploying a machine learning model to automate the classification of surface defects in SRF cavities, crucial for nuclear research advancements. 2. Increased defect classification performance to F1-score of 0.70 using a deep convolutional neural network and transformer encoders, enhancing the facility’s ability to identify and mitigate critical surface defects.3. Processed and analyzed 5,000+ defect images through techniques such as normalization, augmentation, and Fourier analysis, optimizing data quality for model training and improving defect identification insights.

(Also known as The Rembert Group, TN, USA)1. Liquor retail store owners lack data driven information of demand for products which leads to mismanaged inventory. Goal of project was to solve this problem by building machine learning based product demand estimation platform for better inventory management. I implemented and trained ensemble model for product demand estimation for liquor store owners. Achieved mean absolute error of 8 product units for specific product categories.2. Performed data analysis of Iowa Liquor dataset i.e F-test, t-test, Chi-square test, ANOVA test, permutation tests and data visualization. 3. Performed ETL (Extract, Transform, Load) i.e different data sampling methods, feature engineering & selection, normalization, feature encoding, etc. Performed K-fold validation and model optimization with hyper-parameters tuning.4. Performed product clustering using K-means for new products’ matching.

1. Developed a text-prompt based image search tool for 3+ million images using contrastive learning based pretrained model, enabling designers to find sentiment-aligned images for creating effective moodboards.2. Built and deployed data ingestion pipeline for custom data and deployed deep learning models on AWS Cloud, ensuring seamless image search performance.

1. Performed research and implemented multiple object-tracking models for Tennis match videos to detect player, scorebox, ball, net, and court. Achieved >90% test mIOU using Gaussian mixture models, deep convolutional neural networks, and OCR models on videos data.2. Created application for score tracking based auto-trimming of videos to extract rallies.3. Fine-tuned pretrained model for player body segmentation and pose estimation.4. Led data collection and annotation efforts for 100+ videos and successfully deployed models on Linux virtual machines.

1. Developed technology for 3D computer vision based foul detection in Cricket for umpires achieving >95% accuracy for MVP (Minimum Viable Product). 2. Performed research and implemented 3D point cloud based human leg tracking, leg and ground contact detection and depth estimation(resolution=2mm) utilizing Intel RealSense.3. Raised $100K seed investment and received cloud credits worth $120K from IBM GEP for training AI models. 4. Led data collection / annotation team to collect 100K+ images and annotate 50K+ images to ensure data quality for optimal model performance. Led 10 members team to build MVP including industrial designers, CV & Android developer, embedded system engineer, etc.