AI Accelerator

AI Storefront Assistant

Customers often need quick answers about products, inventory availability, order status, shipping estimates, and returns without waiting for customer support. The AI Storefront Assistant provides a conversational shopping experience by understanding customer questions and retrieving live information from IBM Sterling OMS.

Gemini LLMVertex AITool CallingIBM Sterling OMS REST APIsJavaSpring BootReact
Problem

Customers waiting for customer support to answer questions about orders, inventory, shipping estimates and returns — creating friction in the shopping experience and increasing support load.

Solution

Combines Gemini LLM with tool calling so customer messages are processed by the AI model, which invokes predefined tools that call IBM Sterling OMS REST APIs to retrieve live order, inventory and product information. Answers are always grounded in current OMS data, not model hallucination.

Outcome

The prototype enables customers to interact with the storefront using natural language while receiving accurate, real-time information directly from IBM Sterling OMS. Demonstrates how AI improves customer experience without replacing existing OMS business processes.

Key Challenge

Integrating LLM with enterprise APIs while ensuring reliable responses. Implemented tool calling so inventory, order status and fulfillment information always comes directly from OMS — eliminating model hallucination on business-critical data.

Working prototype — AI-powered storefront demonstration AI Commerce

Overview

The AI Storefront Assistant solves a real operational problem: customers need immediate answers about their orders, inventory and shipping without waiting in a support queue. This solution brings generative AI directly into the storefront, connected to live IBM Sterling OMS data.

How It Works

The solution combines a Large Language Model with tool calling. When a customer asks a question, the AI model determines whether external business data is required. If it is, it invokes predefined tools that call IBM Sterling OMS REST APIs to retrieve live order, inventory and product information.

This architecture ensures every answer is grounded in current OMS data rather than the model’s built-in knowledge — critical for enterprise commerce where accuracy matters.

Technical Architecture

The middleware layer sits between the React storefront and IBM Sterling OMS. Gemini LLM on Vertex AI processes natural language input and decides which tools to call. Java Spring Boot middleware handles the tool execution, calling OMS REST APIs and returning structured data back to the model.

Key Challenge Solved

The primary engineering challenge was preventing the model from generating answers from general knowledge when business data is available. Tool calling enforces this — inventory, order status and fulfillment information always comes from OMS, not from the model’s training data.

Status

Working prototype developed as an AI-powered storefront demonstration. Showcases how generative AI can be integrated with IBM Sterling OMS using tool calling and enterprise APIs.

← Back to Projects Work with me