Aug 2023 – Aug 2025 · 2 yrs · New York, New York, United States

Oct 2022 – Aug 2023 · 10 mos · New York, New York, United States

Nov 2020 – Oct 2022 · 1 yr 11 mos · New York, New York, United States

Jun 2020 – Oct 2020 · 4 mos

• Visit sahilkerkar.com for in-depth analysis/findings and github.com/sahilkerkar for source code/relevant datasets
• School Safety: Analyzed the US Department of Education’s School Survey on Crime and Safety from the 2015-2016 school year
• ♦ Part 1: Removed multicollinearity, trained/optimized 8 regression models, and concluded that drug testing, requiring faculty IDs, training teachers on substance abuse, and community involvement in safety were associated with a reduction in violent incidents
• ♦ Part 2: Employed the SMOTE algorithm, trained/optimized 7 multiclass classification models, and determined that a random forest model was most effective at capturing nonlinearity and interaction between features used to predict bullying prevalence
• ♦ Part 3: Implemented dimensionality reduction/clustering to suggest disciplinary action at schools was influenced by student race
• Evolution of Rap: Integrated data from Billboard, Spotify, and Genius APIs and applied NLP techniques to identify trends since 1989
• ♦ Part 1: Visually displayed changes in lyrics to demonstrate that present-day tracks are more profane, shorter, and more repetitive
• ♦ Part 2: Concluded that audio features were useful but not sufficient in predicting if a track would break into the Billboard Top 5

Jun 2019 – May 2020 · 11 mos · New York, New York

• ♦ Utilized SQL, Jupyter, pandas, and time series data to backtest a mean-reversion trading strategy for corporate credit indices, factoring in transaction costs and delays in execution time and concluding that the strategy was profitable in the long run
• ♦ Applied multiple linear regression on cross-sectional data to analyze the option implied volatilities of currencies, ETFs, and government bond futures and provided profitable recommendations for which assets statistically screened expensive versus cheap
• ♦ Built visualizations with matplotlib to convey the relationship between short-term price movements and future realized volatility

Aug 2016 – May 2019 · 2 yrs 9 mos · New York, NY

• ♦ Managed and hedged multi-dimensional portfolio risk by executing broker trades, producing daily risk metrics, and proposing trade ideas based on qualitative macroeconomic views and quantitative statistical analysis
• ♦ Combined Python, Bloomberg Terminal, and Excel to perform time series analysis on the correlation and volatility of various currencies and commodities and presented results to the sales team to drive client engagement and increase revenue

Jun 2015 – Aug 2015 · 2 mos · New York, NY

• ♦ Rotated through the FX options, equity options, and commodities trading desks

May 2014 – Aug 2014 · 3 mos · Bloomfield, Connecticut

• ♦ Reduced model runtime from 600 hours to 1 hour by redesigning and efficiently programming a SAS-based implementation of Monte Carlo simulations to forecast insurance product performance in future fiscal quarters

Aug 2013 – May 2016 · 2 yrs 9 mos · Durham, North Carolina

• ♦ Built a Python tool to clean and apply NLP to over 5,000 transcripts of pharmaceutical company quarterly presentations and determined that the length, tone, and factual density of these presentations have a significant impact on stock price reaction

Aug 2013 – May 2016 · 2 yrs 9 mos · Durham, North Carolina

• ♦ Selected through teacher recommendations and course performance to represent the Duke Mathematics Department and assist 10 to 15 college students with multivariable calculus and linear algebra concepts each week