We investigated the effectiveness of a flow-based approach for code generation using LangGraph, which enables the representation of flows as a graph. By incorporating checks and feedback into the generation process, we found that this approach can significantly improve the accuracy of code execution compared to traditional context-stuffing methods. The addition of reflection and re-try mechanisms allows the model to iteratively improve its answers based on the results of tests and failures, resulting in substantial improvements in code execution relative to a baseline implementation without these features.