Dixit Gajjar

A complete re-implementation of the Momentum Transformer architecture adapted for modern production environments. I refactored the original academic codebase to run on TensorFlow 2.x and engineered a custom data pipeline using Yahoo Finance to replace paid institutional feeds.

Key Research Findings (1990-2025 Backtest):

• Superior Returns: Achieved 1248% Total Return vs 113% for LSTM baseline.
• Risk Metrics: Sharpe Ratio of 2.06 (vs 1.23 for LSTM) with superior Sortino ratio.
• Engineering: Optimized attention mechanisms to train efficiently on local hardware constraints.

Automated Quant Trading System

Python Alpaca API ML GitHub Actions

A fully automated paper-trading engine deployed on the cloud. The system implements a multi-factor strategy combining mean-reversion and trend-following signals.

Key Components:

• Regime Switching: Dynamic exposure adjustment based on Volatility/Trend regimes.
• Signal Processing: Kalman Filters and McGinley Dynamic averages for noise reduction.
• Risk Engine: 15% Annualized Volatility targeting with max-drawdown controls.
• Infrastructure: Serverless execution via GitHub Actions with hourly rebalancing.

Processed high-frequency Trade and Quote (TAQ) data to estimate liquidity metrics, volatility surface, and Probability of Informed Trading (PIN).

Implemented the Longstaff-Schwartz Method (LSM) to accurately price American Options using Least Squares Monte Carlo simulation techniques.

Interactive dashboard that simulates interest rate paths using the Vasicek mean-reverting stochastic differential equation.

Cox-Ingersoll-Ross (CIR) model simulation. Prevents negative interest rates and allows users to visualize term structure dynamics.

Execution strategies and signal generation engine designed for high-frequency trading environments.

Comprehensive risk modeling using Value at Risk (VaR) and Conditional Value at Risk (CVaR) for multi-asset portfolios.

Interactive map visualizing real-time birth statistics across Indian states.