Anika Dhuri

- Designed and analyzed advanced math problems in stochastic processes, probability, statistics, and linear algebra to evaluate and enhance generative AI model reasoning, identifying failure modes in logic and computation. - Evaluated the Deep Research capabilities of SOTA LLMs by assessing ability to conduct literature reviews on financial time series pattern recognition, uncovering reasoning gaps in how macroeconomic indicators (e.g., interest rates, inflation, GDP, unemployment) are used to identify market regimes.

- Improved performance of defensive portfolio strategy leveraging mean variance optimization by creating a neural network to predict beta in event of a market black swan to decrease portfolio losses by ~27%. - Developed a portfolio preselection strategy for S&P 500 stocks utilizing machine learning techniques such as K-means clustering to enhance portfolio diversification (average pairwise asset correlation) by 0.2.

- Worked under mentor in UC Berkeley Economics Department. - Used Monte Carlo simulation to simulate random walk and autoregressive processes, and self-exciting threshold auto-regressive and the smooth transition autoregressive model processes to analyze and visualize behavior of time series decomposition with different filters. - Performed sensitivity analysis for different parameters to empirically determine relationship between the persistence parameter and standard deviation for the cyclical component of a time series decomposition.