Human-in-the-loop
evaluation interface
Merchandisers needed a way to audit AI-modeled search results
Key skills
Opportunity
As H-E-B increasingly relies on predictive and generative AI systems to inform the output of customer search, merchandising teams needed visibility into that output. As H-E-B adjusts to an AI-informed approach in Search, the outputs require human analysis to validate the results or adjust based on business objectives or additional context.
I was tasked with building a Human-in-the-Loop tool to streamline workflows, enhance scalability of search output overrides, and ensure that we're building trust in AI systems and continuing to improve outputs.
Goals
Allow merchants to view search results outputs based on 3 parameters
Allow merchants to change the order of search results, remove an item from a search result, or add an item to a search result
Enable model to ingest merchant's choices to continually learn and improve future output
The approach
Evaluate and map current merchant workflows
Use H-E-B design system, Mortar, as the building blocks of the design
Iterate on designs with support of product design team
Reuse existing and familiar pieces of merchant experience to maintain connection
