Engineering a Property Graph-Powered Architecture for Constraint-Based Recommendations: The SalesBOOST Graph System

The configuration and proposal of industrial appliances during the pre-sales phase is critical: it can be a complex and time-consuming task, complicated by vast product portfolios and rigid technical constraints, but it is decisive to close a deal. Sales teams frequently struggle to identify available alternative products when exact matches are out of stock, largely due to fragmented data across different legacy systems. This thesis presents "SalesBOOST Graph", a constraint-based recommendation system designed to suggest similar products and alleviate these operational bottlenecks. The solution aggregates heterogeneous catalog data into a unified Property Graph within an Oracle Database environment. By leveraging standard SQL/PGQ language, the system applies hard constraints for critical technical specifications, such as voltage and phase compatibility, and soft constraints to calculate a weighted similarity score. Furthermore, OpenAI text embeddings are utilized to refine or rerank candidates based on semantic description similarity. A key requirement and advantage of this constraint-based architecture is its high explainability. Unlike pure black box models, the system provides clear, interpretable reasoning for why specific products are considered similar, a transparency that is further enhanced by interactive graph visualizations in the React frontend. The architecture is supported by a robust FastAPI backend and deployed on Microsoft Azure Container Apps. To support agentic workflows, the system also exposes a Model Context Protocol (MCP) server for seamless integration with Large Language Models. Offline evaluation on a curated dataset demonstrates good ranking performance, confirming the system's ability to accurately prioritize relevant substitute items while surfacing a wide, evenly distributed portion of the product catalog. Successfully deployed in a production enterprise environment, SalesBOOST Graph provides an interpretable, scalable tool that significantly enhances sales capabilities.