Salesforce AI Retail Agent
An AI-powered retail assistant developed through Macquarie University's PACE program with Salesforce as the industry partner, focused on structured product recommendations, customer constraints and cross-sell/upsell workflows using external LLM APIs. An active, in-development project shaped by sponsor feedback.
Project Overview
Problem
Delivered through Macquarie University's PACE program in partnership with Salesforce, the brief was to build a working AI-powered retail assistant that reasons over structured retailer product data rather than acting as a generic chatbot.
What I Built
An in-development AI-powered retail assistant designed around structured product data, customer constraints and recommendation workflows such as product recommendations and cross-sell/upsell, using external LLM APIs. The target architecture is intentionally model- and provider-flexible rather than locked to one LLM vendor. This describes the project's direction and target design; specific components such as tool calling, structured outputs and evaluation are still being built and are not yet fully implemented.
My Contribution
Contributing within a Macquarie University PACE team on the industry-sponsored brief, focused on the retail recommendation logic and on keeping the architecture provider-flexible rather than tied to a single LLM vendor.
Engineering
Architecture & Decisions
Planned architecture: user request, then intent and constraint extraction, then tool calling (product search, filter by price, check availability, compare products, find alternatives or complementary products), then recommendation logic, then a structured response. This is the current target design; not all stages are fully implemented yet.
Tech Stack
Testing & Delivery
AI evaluation and testing practices are a current focus area as the implementation matures; not yet fully in place.
In development; not yet deployed or released.
Outcome & Current Status
Stakeholder & User Impact
Delivered with Salesforce as the industry sponsor, with sponsor check-ins and iterative feedback shaping scope and direction.
Current Status
Active, in-progress industry project. Capabilities such as tool calling, structured outputs, evaluation and guardrails are current or planned work rather than completed features, and are only described as done once implemented.
Future Direction
Planned next milestones: implement and validate tool calling for product search and filtering, structured output handling, and AI evaluation/observability.