coshop
coshop is a benchmark for evaluating conversational shopping assistant agents that must elicit latent user preferences through dialogue and retrieve the correct item(s) from a catalog.
Subpackages
| Package | Description |
|---|---|
coshop.data |
Dataset loading, item representation, and utility scoring |
coshop.evaluation |
NDCG-based metrics and per-turn budget tracking |
coshop.tools |
LangChain-compatible retrieval and history-search tools |
coshop.user_simulator |
Simulated users that reveal preferences progressively |
coshop.utils |
Shared data types, agent base classes, and helper utilities |