A massively parallel research and production pipeline engineered to compute, orthogonalize, and backtest thousands of cross-sectional quantitative alpha expressions simultaneously across global asset universes.
Engineering Highlights:
- Custom CUDA Graph Execution: Fused tensor operations eliminating host-device memory transfers for rolling statistical windows.
- Gram-Schmidt Factor Orthogonalization: Parallel GPU matrix factorization ensuring zero collinearity with common risk factors (Fama-French 5-Factor / Barra).
- 100x Speedup Over NumPy/Pandas: Evaluates 15 years of tick-level tick data in under 12 seconds.