Developing various tools for pricing Credit derivative products by directly using the C++ analytics library in Python. The classes and functions of the analytics library (written in C++) are exposed to Python through an in-house interoperability library. This has enabled us to quickly prototype new models and roll out new features for users. Following are a few of the Python based tools I've developed:1. CDS Option Pricer - Prices options to enter a CDS contract using Black Scholes Model - Developed an intraday scratch pricer for bespoke trades and also an eod risk viewer - Implemented model to price options in future using the latest market data2. Credit Curve Calibrator - Takes CDS spreads of a reference entity from the market and calculates its hazard rates3. CDS Index Tranche Pricer - Developing a tool to price CDS Index tranches using Gaussian Copula Correlated recovery (GCCR) model

Motivated by the mode estimation problem of an unknown multivariate probability density function, westudy the problem of identifying the point with the minimum kth nearest neighbor distance for a given dataset of n points. We study the case where the pairwise distances are apriori unknown, but we have access to an oracle which we can query to get noisy information about the distance between any pair of points.- Designed a 2-layer sequential multi-armed bandit algorithm to find the point with minimum k-NN distance- Analyzed the performance of the proposed algorithm for two types of oracles: (i) oracle returns the distance of two points along a random dimension, and (ii) oracle adds a sub-Gaussian noise to the true distance of two points- Proposed an Information Theoretically optimal algorithm that estimates the kth nearest neighbor of a point- Showcased optimality of the algorithm by finding its upper and lower bounds and proving they are of same order

- Designed the global minimum variance and tangent portfolio consisting of 8 stocks with and without short-selling- Attained a return of 60.88% with a risk of 0.42 and a Sharpe ratio of 1.42 for the tangent portfolio with shorting- Employed Fama-French and Capital Asset Pricing models to interpret dependence of return on risk for portfolios- Computed a portfolio of Danish Bonds with a pre-specified duration using a Nelson-Siegel term structure model

Visual compatibility prediction refers to the task of determining if a set of clothing items (an outfit)go well together. There are three modalities involved in modeling an outfit: category (jeans, shirt, shoes, etc.) of individual clothing items, the context of an item (a set of other items it is compatible with), and the fashion style of an outfit. We propose a unified framework combining all of these modalities.- Outperformed the current state-of-the-art model by 7% in measuring the compatibility of a set of clothing items- Introduced a category conditioned Graph Convolution Network to model the category and context of the items- Developed an Attention based Autoencoder for clustering the outfits in 6 clusters based on their fashion styles- Used a Reinforcement Learning technique to combine the two measures further improving the accuracy by 1%

Developed an in-house app analytics service for OkCredit, a startup which provides a mobile-based digital ledge- Designed an infrastructure to collect the user interactions of 10k+ users from a mobile app for product analytics. Built a stateless server in Google Go to store data in a Cassandra database and transfer it daily to Amazon S3. Created an Android Library to store user interaction data locally on the mobile phone and send it to the server.- Incorporated Google Sign-In as an ID provider in an Oauth 2.0 protocol based authentication service in Google Go.- Devised and performed unit & load tests of REST APIs to calculate their maximum load as a function of resources