• Designed, developed, tested, and deployed Natural Language Processing (NLP) models such as Sentence BERT transformer model, to provide job recommendations based on candidates' previous applications.
• Implemented job matching process utilizing NLP techniques to identify related jobs based on job description sentence embeddings and density-based spatial clustering.
• Enhanced matching accuracy by incorporating a variant of the Sentence BERT transformer model and fine-tuning the model using seniority labeled text in job descriptions to match jobs of the same seniority level.
• Achieved a significant 300% increase in monthly job applications as a result of the job recommendation system implementation.

This is a study done with Prof. Sadka, where we try to understand the abnormal returns caused by social media activity in the short-term. In this study, we:

• Developed firm-level indicators using Natural Language Processing (NLP) techniques on unstructured daily data from 300 social media sources.
• Constructed weekly rebalanced value-weighted and equal-weighted portfolios based on these indicators, generating annual alphas of up to 50% for the year 2020-2021.
• Demonstrated significant predictive power of returns for the social media-derived activity indicators based on Fama-Macbeth regressions, which involves estimating time-series regressions for each cross-sectional period to examine the average relationship between independent and dependent variables, and its statistical significance.
• Extracted abnormal returns through Fama-French factor models, which were not explained by the conventional risk factors that financial assets are exposed to.

• This paper is a cross-country study where the mutual fund holdings are used to understand the mechanism for change in Betting-against-beta (BAB) strategy returns.

• This study builds upon the earlier published work in the Review of Financial Studies (RFS) in 2020 by my co-authors (Prof. Hoberg, Prof. Kumar, Prof. Prabhala).
• We examine how many peer funds are required to uncover managerial skill, and what dimensions are useful for benchmarking
• We incorporate customized benchmarking of global mutual funds in a time-varying factor-premium framework.

• Proposed time-varying variances adjusted to skewness and kurtosis (GARCH-SK), dynamic conditional correlations (DCC-GARCH) and credit ratings as combined proxies for risk
• Performed a non-linear multi-objective optimization to obtain the pareto optimal front using Non-dominated Sorting Genetic Algorithm (NSGA-II)

• Studied volatility spillovers in metal prices of Indian commodity market (MCX-India) from global commodity markets viz., LME, CME/ICE, EUREX, SHFE using DCC-GARCH and BEKK-GARCH
• Calculated time-varying hedge ratios for each of these metals

• This paper studies the cross-sectional returns of the Indian mutual funds, which are known to outperform their benchmark indices
• We find that all factor alphas are significant in the rolling 5-year windows followed by the 2008 Financial crisis.
• We also find that domestic funds on an average generate 6% more returns than foreign funds