PhoneMall

The problem

Traditional catalog filters make customers understand dozens of product parameters before they can choose. A generic chat model has the opposite problem: it may invent products, prices, or stock. PhoneMall places the full commerce workflow and a RAG shopping assistant in the same system, grounding recommendations in the current catalog.

Solution

The system retrieves candidates from products that are currently available, then gives those candidates and the user’s constraints to the model. Responses stream through SSE. The model helps explain trade-offs, while product, inventory, and transaction state remain under the control of the business services.

Business workflow

• Customers, products, SKUs, cart, orders, inventory, payment, cancellation, and refunds.
• Merchant product management, fulfillment, and AI configuration.
• Spring Cloud service governance, MySQL persistence, Redis caching, and self-hosted Docker deployment.

Verification

The review snapshot dated 2026-07-15 contains four backend services and 39 backend tests. Repository CI also checks the backend, frontend build, Compose and Nginx configuration, source scanning, and container image security. The source repository is now public, so readers can inspect the implementation and release history directly. Because the repository has no selected open-source license, the site labels it a public repository rather than open source.

Trade-offs and limitations

The project is still in Alpha. In-memory vector search is best suited to small and medium catalogs. Frontend end-to-end coverage, retrieval evaluation, realistic load baselines, and a consistent demo media package still need work. This page does not claim that the system is a public demo or a production-ready service.

Next steps

• Add sanitized interface captures, an architecture diagram, and an end-to-end demo video.
• Establish retrieval evaluation and performance baselines.
• Expand frontend end-to-end tests.
• Evaluate a persistent vector index.