The shopping agent was built to provide reliable product recommendations by ensuring every suggestion meets the user's requirements, including price, category, and availability. Through competitive evaluation against standardized shopping tasks, the team discovered that a...
Improved an existing shopping agent by treating the independent grading criteria as its specification. The process involved rebuilding its prompts around measurable requirements, testing one change at a time, and selecting GPT-6 (Astra) over Opus 5.1 based on actual graded...
