Traditional quantitative factor discovery relies heavily on manual processes where analysts translate academic insights into alpha signals through labor-intensive workflows. While recent advances in large language models (LLMs) show promise for automating idea generation, simple prompt-based approaches often produce redundant, invalid, or economically unsound factors. We introduce AlphaMind, a closed-loop system that combines LLM-driven creativity with rigorous empirical validation and structured knowledge management. At its core, AlphaMind maintains a Dynamic Factor Repository storing executable code, natural language explanations, and standardized formulas for each validated signal. The system operates through iterative discovery cycles: an Alpha Generator produces new factor ideas using retrieval-augmented generation from financial literature, a translation module converts these ideas into executable code, a Backtesting Module filters candidates through rolling-window correlations and cross-sectional tests, and a Second-Pass Reasoning LLM ensures economic soundness and formula simplicity. Only factors passing all validation stages enter the repository, forming an expanding knowledge base for future generations. This architecture transforms quantitative factor mining from ad-hoc experimentation into a systematic, self-improving research engine. We demonstrate that AlphaMind can discover novel, economically grounded factors while maintaining the transparency and reproducibility essential for institutional portfolio construction.

Li Honglin is a Ph.D. candidate in the Department of Industrial Systems Engineering and Management, National University of Singapore. His research focuses on the intersection of finance and technology, specifically in quantitative finance, statistical arbitrage, token economics, and blockchain security & technology. He is passionate about applying mathematical and computational methods to solve complex problems in financial markets and emerging blockchain ecosystems.