Jan 2020 – Oct 2022 · 2 yrs 9 mos · Bengaluru, Karnataka, India

Oct 2015 – Aug 2019 · 3 yrs 10 mos · Noida Area, India

• Click through rate (CTR) prediction for Advertisement serving using probabilistic machine learning models and deep learning techniques.
• Contextual recommendation service using the word embeddings provided word2vec, bert, elmo to get low dimensional vector space representation of users click history.
• Item recommendation for interest based recommendation using ALS variant of collaborative filtering for different publisher-platform combinations.
• Proof of Concept on CTR prediction for the AdImpressions using wide and deep model of tensorflow/keras. Took the embeddings for the categorical features to find the semantic relationship in terms of the geometric distance between vectors.
• Handling the data layer at Times Internet. Apache Kakfa is used for the storage of raw logs from the AdServer and Spark Streaming Jobs are written to perform complex real time aggregations for dashboards and finding interesting insights from the data which helps in decision making process for advertisement serving.

Aug 2014 – Sep 2015 · 1 yr 1 mo

• Research and Development of analytical operations on distributed framework like Hadoop (YARN), Spark (Version 1.2.0). Exploring the data science operations and their optimized parallel execution over fast data processing framework Spark. This involves the data preparation steps like Missing Value Treatment, Outlier Detection, and Binning etc.
• Development of R based GUI tool (Shiny package) which aimed at automating data science operations and statistical modelling techniques. This tool allows analysts to perform the operations in an intuitive and graphical manner without any manual intervention.
• Understanding the fundamental concepts of Machine Learning and Statistics to implement them in the above tools

Sep 2012 – Jul 2014 · 1 yr 10 mos · Noida Area, India

• The Private Wealth Management division of Morgan Stanley is managed using a product called GIM2 written in C++/ESQL/Informix/Solaris used for reporting purpose for PWM clients. EMS (Error Management System) is an integral part of this system to ensure quality of data along the subsystem. It is responsible to highlight data issues in Gim2 and to correct them by using an automated solution if possible. The EMS GUI is used by the business users to fix the daily breaks in production environment interacts with the backend using the EMSserver.
• ROLES & RESPONSIBILITIES:
• Understanding the requirements from the business users, feasibility, implementing the new functionalities and deploying them into production environment on time
• Development of new error modules for data quality checks
• Requirements capture, design, troubleshooting, implementation and coordination of testing with the business users and the QA teams
• PWM Amortization
• The project is aimed to bring the Amortization Factors Information from a third party Wall Street Concepts, in order to get the correct Amortization Values Calculated for Client Statements. This Information is then supposed to automate the process of calculation of the changes in the positions of the clients holding the securities.
• The system is designed to transmit the Transaction/Security and Call Schedule information from Morgan Stanley to Wall Street Concepts which then resends the amortization factors. Automated File Watchers were also written that would get triggered and load data in GIM2 when we received back the factors from WSC.

Sep 2009 – Jul 2012 · 2 yrs 10 mos · Hyderabad Area, India

• Addition of new features to the Target Services Library as per project requirements.
• Providing functionality for controlling concurrent execution of finite state machines, for delivering messages (both application internal and system wide, and for providing timing, debugging and logging services (Trace and Error).
• Service library is responsible for creation and destruction of capsules, and delivery of message between the capsules.
• The Target Services Library was updated with UML Signal History for effective debugging. The main aim was to dump the last sequence of signals exchanged into a message queue with all the information. Debugging commands were also introduced for run time debugging for both the application and controller level.

Jun 2008 – Jul 2008 · 1 mo · Kalamazoo, Michigan Area