Multi-Agent LLM System for Automated Neural Network Tuning

A multi-agent framework utilizing analyst, coder, and recorder roles automates the tuning of specific neural network components within a sandboxed environment. The system successfully optimized hyperparameters for spiking neural networks and Physics-Informed Neural Networks (PINN) by generating targeted code rather than full architectures.