Technical Summary

• Built a real-time ingestion + feature engineering pipeline from FRED macro/market series (rates, spreads, volatility, inflation proxies, equity signals).
• Transformed raw inputs into rolling z-scored structural factors (yield-curve slope, spread pressure, volatility intensity, correlation concentration) for model stability.
• Estimated latent macro states with EM-fitted Kalman filtering in linear state-space form, exposing hidden stress/liquidity/growth/inflation dynamics.
• Ran Numba-accelerated Monte Carlo simulations with regime-switching Markov transitions to compare policy scenarios (Ease/Hold/Tighten) over configurable horizons.
• Implemented a PPO baseline in a custom Gymnasium environment (actor-critic + GAE) to benchmark learned rate adjustments against analytical recommendations.
• Defined a risk-aware objective that jointly penalizes stress, expected shortfall, and crisis probability, improving decision transparency under tail-risk conditions.