I am a Software Development Engineer at Reliance Jio Cloud Services. Not long back, I graduated from the Indian Institute of Technology, Guwahati with a Bachelor's degree (B.Tech) in Mathematics & Computing.
I am an avid traveller. I have visited most of the European countries. In India, I had the opportunity to live an amazing 4 yrs in the campus of IIT Guwahati. Here is the pic. I also love to read books particularly non-fiction stuff and I can binge watch TV series at will!
Software Development Engineer, Reliance Jio Cloud Services Feb 2016 - Present
The research problem that my work addresses is to predict the hardware failures before they occur. Predicting hardware failure in advance is important because failures cost a lot of money sometimes more than false positive. I have developed a monitoring framework that collects the system metrics from 1000s of servers. One of the challenges of predictive maintenance is the imbalance in the dataset between successes and the failures as there are so few failures. For instance, after monitoring fans of all servers for around 5 months, there has been less than 30 fan failures. To resolve this, I used SMOTE (Synthetic Minority Over-sampling) rather than a normal minority over-sampling for it forces decision region of minority class to become more general. Also, hardware failures are usually dependent on many attributes. For example, a fan failure depends on ambient temperature, location, manufacturer, model, age, cpu load, network load, disk load etc.
Software Engineer, Samsung Research India Bangalore July 2015 - Jan 2016
Worked on developing Cloud services for Internet Of Things (IoT) in B2B domain. Technologies used in the project being Spring-Java, Kafka, Storm etc
Undergraduate Research Assistant, Dept. of Finance, Ulm University Summers 2014
Supervisor: Dr. Gunter Loeffler, Dept. of Finance, Ulm University
Research at Ulm University involved determining the ways in which the net negative sentiment of a Moody's investor report affects the future rating changes of the firm. Another aspect that had to be taken into consideration was the metadata of the report such as analyst names, designation, additional qualification etc. To move forward, we analysed the tone of reports using bags of words approach. The metadata and net negative sentiment of each report were stored in a relational database. Finally, to train the model for prediction, we used logistic regression because feature space was not large. Moreover, the main role of this exercise was to classify the future rating changes as positive or negative. Of the many interesting results, we found out that a company has more probability to be downgraded if it has high analyst coverage. We also found out interestingly, that there was a surge in analyst hiring during the 2006-2007 (just preceding financial crisis of 2008). Moreover there was a peak in number of companies rated per analyst during this period.
Modelling temperature and pricing of weather derivatives for the Indian markets Fall 2014 - Spring 2015
Advisor: Dr. Siddhartha P. Chakrabarty, Department of Mathematics, IIT Guwahati
I chose this topic specifically as the derivatives market in India is still in its infancy and weather derivatives would be a norm for India in future as it still depends on agriculture.
The weather derivatives were priced for 6 different Indian cities and I used two methods namely the closed form formula and Monte Carlo simulation to determine the price of weather derivatives. The prices determined by two methods were approximately same.
Emergency Blood Management Spring 2014
As part of the project, a MySQL database was developed which had the information of donors, blood banks who were ready to donate blood in cases of utmost emergency. Interface was designed using HTML. PHP was used in backed to interact with the database.
Simulating Fractional Brownian Motion to estimate the implied volatility Summers 2013
As a contributor to the project, my work involved simulating the fractional Brownian motion through R and applying statistical model to estimate the implied volatility.