This research introduces a novel ensemble approach that combines outputs from multiple LLMs to generate more reliable code solutions, rather than relying on a single model's output.

• Creates multiple candidate programs from different LLMs
• Applies structured voting mechanisms using syntactic and semantic similarity metrics
• Selects the most reliable solution through consensus
• Improves robustness, accuracy and generalization in code generation tasks

This innovation addresses critical engineering challenges by reducing potential bugs and vulnerabilities in AI-generated code, providing a practical method for enhancing code quality in development workflows without additional manual intervention.

Enhancing LLM Code Generation with Ensembles: A Similarity-Based Selection Approach