GPU-accelerated quantitative derivatives engine: FiLM-FNO rough volatility calibration (<1ms), multi-fidelity PDE ensemble autocall pricing (0.99 bps RMSE, SR 26-2 risk guardian), deep hedging, and Next.js/FastAPI streaming (1,126 tests).
nextjs cuda pytorch quantitative-finance tensorrt derivatives-pricing fastapi volatility-surface deep-hedging fourier-neural-operator rough-heston autocallables
-
Updated
Oct 18, 2026 - Python