Cart Assistant: Agentic Grocery Shopping on Uber Eats

Summary

Cart Assistant addresses grocery requests that begin as lists, recipe images, or vague plans and turns them into draft carts shoppers can review and edit. It uses a multi-prompt state graph that assigns planning, candidate retrieval, semantic relevance, constraint enforcement, quantity selection, guardrails, assembly, and content refinement to focused stages. LLMs handle ambiguity and structured outputs, while backend APIs, deterministic arithmetic, validation, optimization, and aggregation manage inventory, pricing, eligibility, sellable units, and cart construction. The design separates retrieval language from reasoning context, preserves quantity provenance, supports parallel item processing, and uses deterministic checks plus an LLM judge against synthetic and production-derived cases. According to the conclusion, this decomposition makes the system more controllable and inspectable while producing useful draft carts without removing shopper review, editing, and decision-making.

Takeaways

1. 1

   Retrieval language and reasoning context are stored separately, allowing concise inventory queries while preserving details such as “gluten-free is a strict user constraint” for later decisions.

2. 2

   Quantity resolution combines LLM reasoning about human expressions and packaging with deterministic arithmetic and validation, while preserving whether a quantity was explicit or system-estimated.

3. 3

   Evaluation combines rule-based verification with a human-calibrated multimodal LLM judge across synthetic requests and anonymized production-derived edge cases.

FAQ