Enhancing LLM performance with reasoning using deterministic feedback loops