001 : Real-Time News Sentiment Signal Engine for Trading

  • Designed and implemented a real-time sentiment engine that processes financial news and social media headlines to generate trading signals. Used FinBERT to classify market-moving sentiment and computed rolling sentiment deltas to trigger long/short alerts on tickers like BTC and TSLA. Integrated with NewsAPI and Twitter (free-tier) for live data. Final outputs include classified news, signals, and a CSV-based backtest-ready format.

    • 002 : VaR-CVaR Estimation for AI Stocks and Bitcoin Correlated Portfolio

  • Developed a robust Monte Carlo simulation framework to compute Value-at-Risk (VaR) and Conditional VaR (CVaR) for a portfolio blending AI equities and Bitcoin. Modeled dependencies using t-copulas and preserved correlations via Cholesky decomposition. Simulated 100,000+ return paths to evaluate portfolio risk under various horizons and confidence levels. Output includes tail risk metrics and distribution plots.

    • 003 : Counterparty Credit Risk Model using CVA and DVA Adjustments

  • Built a credit risk engine to estimate CVA and DVA for OTC derivatives like FX forwards using simulated exposure paths and bootstrapped survival probabilities. Exposure profiles were generated using Geometric Brownian Motion. CVA and DVA were computed through numerical integration using loss-given-default and risk-free discounting. Outputs include EE, EPE, ENE charts and credit-adjusted valuation metrics.

    • 004 :Value-at-Risk (VaR) Estimation Model for Multi-Asset Portfolio

  • Created a modular risk framework to estimate VaR using Historical Simulation, Parametric Gaussian approximation, and Monte Carlo methods. Applied these methods to a diversified portfolio of SPY, GLD, TLT, and BTC, and calculated VaR over multiple time horizons and confidence levels. Delivered interpretable outputs through loss distribution plots and tabular summaries.

    • 005 : AI-Powered Intraday Market Making Strategy for Bitcoin Futures

  • Implemented a Deep Q-Network (DQN) agent to perform intraday market making in a simulated Bitcoin limit order book environment. The custom Gym-compatible environment modeled price evolution, inventory control, and fill probabilities. The RL agent learned to optimize quoting decisions by maximizing PnL while minimizing inventory risk. Performance was evaluated using PnL and inventory plots.