Design And Implementation of an AI-Powered Chatbot for Procurement Management Systems Using Laravel, GPT, And Pinecone – A Case Study

Abstract

Background: The integration of AI-powered chatbots into enterprise systems, particularly in sensitive domains such as procurement, presents both opportunities and challenges. While conversational AI can enhance accessibility and efficiency, direct interaction with databases through natural language queries introduces significant security risks, including SQL injection and potential data exposure. There is a growing need for architectures that balance usability, performance, and ethical compliance.

Methods: This study employed a Design Science Research (DSR) methodology involving iterative design, development, and real-world evaluation. The system was developed using Laravel, MySQL, OpenAI GPT models, and the Pinecone vector database. An initial architecture based on direct SQL query generation was replaced with a Retrieval-Augmented Generation (RAG) approach, where only trace identifiers were embedded in the vector database. This design enabled controlled data retrieval while preventing direct exposure of backend systems.

Findings: The redesigned RAG-based system demonstrated substantial improvements in both performance and security. Results showed a 62% reduction in backend query load, an average response time of 1.2 seconds, and a trace retrieval accuracy of 93%. User satisfaction was also high, indicating effective real-world applicability. The findings confirm that separating language models from direct database access enhances operational efficiency while mitigating security risks.

Conclusion: The study demonstrates that secure architectural design-specifically the adoption of RAG—can enable the responsible deployment of AI chatbots in enterprise environments. By embedding principles such as data minimization, transparency, and regulatory compliance into system design, organizations can achieve both efficiency and trustworthiness. Future research should focus on scalability, customization, and reducing reliance on external AI services.