AlphaRefinery: LLM-assisted alpha-factor research
I built AlphaRefinery, an LLM-assisted platform for structured alpha-factor research that generates economically interpretable candidates by recombining market features and operators across economic themes. I designed the factor registry, staged search, and experiment memory to support iterative generation, evaluation, and refinement. UCB-style exploration and MMR reranking balance candidate quality and diversity, while evaluation considers predictive power, transaction costs, robustness, and redundancy with existing factors.