Show HN: LoongFlow – Directed evolutionary search framework for LLM agents (github.com)

LoongFlow has launched an innovative framework for developing self-optimizing AI agents, emphasizing modularity and evolutionary construction. This framework represents a significant advancement in the AI/ML community by automating the optimization tasks for algorithms and mathematical problems through a self-evolving machine learning agent that comprehends data, constructs models, and delivers solutions autonomously. Rooted in the philosophical principles of Wang Yangming, LoongFlow promotes the integration of knowledge and action, leading to a paradigm shift from passive tools to autonomous intelligence. The framework employs a novel PES (Plan-Execute-Summarize) evolutionary model, which enhances efficiency by approximately 60% compared to traditional methods. This advancement minimizes issues associated with randomness and low-quality generations while boosting stability and reproducibility due to its systematic architecture. Notably, LoongFlow outperformed previous benchmarks in tackling complex mathematical problems and achieved gold medals in numerous Kaggle competitions, demonstrating its practical effectiveness. With its Python 3.12 requirement, LoongFlow is poised to empower developers to create high-quality evolutionary agents with reduced development costs and enhanced application flexibility.