Ravi Choudhary Email & Phone Number

San Francisco Bay Area

Working on DNN algorithms for improving recommendation systems for financial products.

Worked on creation and maintenance of recommendation system algorithms for consumption and creation of Karaoke performances on the app. Worked on improving candidate generation using Word2Vec in addition to Factorization Machine.

New York City Metropolitan Area

Worked on processing PDF documents, natural language query answering models on tables, few-shot classification and unsupervised learning on text.

San Francisco, California

Developed an ML model to predict Free cash of a customer. This model is being used to target customers to improve the ROI of marketing campaigns by targeting customers with higher chances of approvability.- Created an ordinal classification model using LightGBM. The model had a correlation of 0.52 with true results compared to a benchmark of 0.38. Automated the model using AWS Airflow. Compared the performance of LightGBM with a Feedforward network with Batch Norm and Self-Normalizing Neural Network.Created an uplift model to identify prospective customers who would respond to multiple mails compared to just one mail.

Created an ML model to predict if a customer will fail to pay all or part of their credit card bills within the next few months. The model was used to reduce exposure of the bank to the customer. - A combination of XGBoost and Logistic Regression based classification model was built. The implementation of the model resulted in a net credit saving of USD 7 Million annually. - Worked on improving the model through new features based on several case reviews of credit defaulters.- Studied the various use cases of the ML model and created a methodology to link the performance (accuracy deviation) of the model with the bottom line of the company. The methodology is currently used to alert the stakeholders of the monetary risk due to deterioration of the model performance.- Worked on reducing the impact of the ML model on newly acquired customers (through portfolio acquisition from a competitor) with missing data.Created an ML model to predict if a payment made by a customer for a credit card bill through check will fail in the future.- An XGBoost based classification model was built. - Worked on complex feature engineering to capture the payment and spend pattern of the customer. Features led to 14 bps lift in the Gini coefficient of the model.Worked in a 3 member team to create an ensemble model of Random Forest and KNN to assign an initial line to a credit card customer which will result in maximum profitability.

Gurgaon, India

Provided analytical support and strategic consulting as a part of the Fraud Analytics - Loss Mitigation Team and Identity Theft Analytics Team for the credit card business of a major US Bank. Responsibilities included• Harnessing large volumes of transaction/application data, conducting complex analyses and developing fraud rules and strategies aimed to minimize losses, customer inconvenience and possible fraud. Created fraud management strategies which increased the expected yearly EBIT by $ 3 MM. • Monitoring and evaluating performance of existing fraud strategies in order to improve efficiency and effectiveness. Evaluated performance of 1000+ rules/strategies for identification and removal of strategies with low performance which resulted in an increase of 10% in the overall efficiency of the system.• Analyzing spend and fraud trends in order to recommend new variables to enhance fraud detection and minimize customer inconvenience.• Recommending strategy adjustments based on emerging fraud trends.

Kharagpur

Collaborated with a team of 40 members, Managers, locals of Gopali, students and professors from IIT for smooth functioning of Jagriti Vidya Mandir, a school run by GYWS for underprivileged kids in Kharagpur. Supervised a team of 5 members for regular website maintenance.

Kharagpur

Responsible for training a team of 5 members in the field of website creation and maintenance.

Kharagpur

Co-developed and handled website of Kshitij, Techno-management fest of IIT Kharagpur. Co-led web team of 10 members to manage various online events of Kshitij 2013.

Kharagpur

Developed websites for Cryptex and Disha, online events organized in Kshitij 2012. Recruited and trained a team of 10 sub-heads from among 30 potential candidates for Kshitij 2013.

Delhi

Performed an in-depth analysis of Police-Public interface in India through various reports, surveys, interviews, case studies. Came up with a list of good practice guidelines and recommendations to eliminate the lacuna in the present police-public interface across India.

Created and managed the initial Facebook page of Buyhatke. Presented initial idea of BuyHatke at the finals of Concipio, Global Entrepreneurship Summit 2012.

Master Of Science - Ms, Data Science, 3.96/4

Activities and Societies: Teaching Assistant for Machine Learning, Deep Learning in Medicine, Statistical Computing, Data Science.

Dual Degree (B.Tech + M.Tech), Mechanical Engineering, 8.62/10

Activities and Societies: Kshitij, GYWS