QuantOpt - AI Portfolio Optimization

Overview

A comprehensive machine learning system for portfolio optimization using PySpark and PyTorch. Implements Modern Portfolio Theory optimizers, Deep Reinforcement Learning (DDPG/PPO) for dynamic rebalancing, and Monte Carlo risk simulations with robust backtesting and evaluation.

Challenges & Solutions

Designing a unified pipeline that scales from classical optimization to deep RL while ensuring reproducibility, robust feature engineering, and realistic backtesting with transaction costs and stress scenarios.

Technical Achievements

• End-to-end pipeline: preprocessing, feature engineering, modeling, and evaluation with PySpark + PyTorch
• Multiple optimizers: MPT (max Sharpe, min variance), Deep RL rebalancing (DDPG/PPO), and Monte Carlo risk
• Backtesting framework with stress testing (e.g., 2008-like crash scenarios) and rich metrics
• Config-driven architecture with reproducible runs and scalable data handling
• Extensible module layout with unit tests and results reporting

Technologies Used