Hubble framework uses large language models to automatically discover profitable trading factors while avoiding overfitting and improving interpretability

Hubble is a closed-loop factor mining system that leverages LLMs as intelligent search heuristics to discover alpha factors in quantitative finance

The framework constrains LLM outputs using a domain-specific operator language and AST-based execution sandbox to ensure safe and valid factor generation

Candidate factors are rigorously evaluated through cross-sectional Rank Information Coefficient (RankIC), annualized Information Ratio, and portfolio turnover metrics

An evolutionary feedback loop returns top-performing factors and structured error diagnostics back to the LLM for iterative refinement across multiple generations

This approach addresses limitations of existing genetic programming methods that often produce complex, uninterpretable formulas prone to overfitting