This study explores a new method for improving financial fraud detection by combining advanced natural language processing (NLP) with graph database technologies. By using there Llama2 NLP model, the proposed research will analyze the unstructured financial data to find the difficult hidden patterns that are linked to the fraudulent behaviour. We use Neo4j,a strong graph database which is used for mapping and examining the complex relationships in between the financial elements, finding the hidden anomalies that the existing systems may miss. This integrated approach allows more effective analysis of both the structured and the unstructured data, improving the fraud detection performance in terms of both the accuracy and the speed. And also this method addresses the main issues like high rate of false positive rates and limited and less scalability which is generally found in the existing detection models, introducing a new solution to enhance financial security and efficiency.