This project is an experiment-first implementation of level-2 market making, treating quoting as a coupled control problem: choose bid/ask placements while managing inventory evolution and fill uncertainty.

The project follows established microstructure formulations -- spread/LOB modeling in the spirit of Guilbaud-Pham and inventory-risk control in the spirit of Gueant-Lehalle -- then maps those assumptions onto a real message-driven engine to observe where theory breaks or needs correction.

Quote placement is driven by order-book state and a noise/price-vibration view (spectral analysis), while inventory is actively skewed using tail-aware risk measures (Cornish-Fisher-style) and latency-aware fill models. The stack was built as model, simulator/engine, live-demo validation, emphasizing execution realism (throughput, latency, queue effects) rather than purely offline backtests.