regime

Using multiple time series observations, split time into regimes using an array of representation learning techniques with explainability and robustness techniques for the transitions

From-scratch 2-state Gaussian HMM on 16y SPX realized vol — regime-tilted overlay lifts Sharpe 0.75->1.08 (+43.9%) and cuts max DD -33.9% -> -19.4%. No hmmlearn/sklearn; pure numpy forward-backward + Baum-Welch.

RAMME: Market Microstructure Engine Real-time market microstructure simulator with regime-based ticks, live plotting, and PnL tracking.

MCP server for real-time market regime classification — give AI agents market awareness. SPY regime detection, VIX analysis, trading implications.