JP Morgan

Attention & Deep Learning Based Email Classifier

• Service & Implementation team had to go through 100s of incoming eMails in a day to comprehend and tag relevant team for issue resolution
• Implemented Attention based neural network using Tensorflow to understand and classify emails into subsequent categories and teams
• Delivered solution reduced manual labor required to identify and tag emails to teams, for more than 150 mailboxes
• Tech Stack Used: Python, Tensorflow, JupyterLab, AWS

Classification Based Analogous Client Finder

• Currently numerous billing platform are used by JP Morgan's commercial banking unit, tailoring to client's specificity which adds extra burden on operations and maintenance cost
• Proposed XGBoost based classification solution lead to operational savings in production deferral costs through strategic client migration
• Provided solution was used to segregate for migration plan
• Tech Stack Used: Python, Tableau, Scikit-Learn

Latent Dirichlet Allocation Based Request Categorization

• Eliminated Onboarding & Servicing team's invisibility to unseen trends and change in key issue subjects
• Developed Latent Dirichlet Allocation based topic modeling & pattern recognition, leading to automated category and subject identifier
• Transformed solution was adopted by Commercial Banking's client service team to understand change in client issues to improve process optimization
• Tech Stack Used: Python, NLTK, Sklearn

Mu Sigma

Random Forest Based Manufacturing Halts Reduction

• Assisted an aluminum conglomerate to reduce unplanned maintenance shutdown, production halts and improve equipment life cycle
• Proposed Random Forest based predictive solution lead to operational savings of over $30MM annually in production deferral costs
• Tech Stack Used: Python, Scikit-Learn, Azure DataBricks, Tableau

Computer Vision & Deep Learning Based Brick & Mortar Customer Analysis

• Understanding offline customer's behavior patterns to make better decisions in store operations (staff management, product placements, etc)
• Implemented in-store video analytics solution using Single shot multibox detector (SSD) based YOLO V3 solution
• Delivered hour level data about customer entry, exit & in-store count, number of aisle visits, traditional checkout counter count, Scan-&-Go counters encouragement and others
• Tech Stack Used: Python, YOLO v3, OpenCv

Natural Language Processing Based Early Trends Detector

• Eliminated sourcing & procurement team's invisibility to unseen trends
• Developed Natural Language Processing based model lead to 3 fold decrease in Out-of-Stock scenarios
• Transformed solution was adopted by clients as their official banner product for 2019 Black Friday Sale
• Tech Stack Used: Python, NLTK, Tableau