WUFUS(CART) - E-Commerce Shopping Cart Assistant

wufus-CART-8B is an 8.2 billion parameter model developed by owlgebra-ai, based on the Qwen3-8B architecture. It has been fine-tuned using on-policy DAPO training (Reinforcement Learning) with OpenEnv, specifically for handling complex, multi-turn e-commerce shopping conversations that leverage external tools. The model's primary function is to act as an intelligent shopping cart assistant, guiding users through product discovery, variant selection, and cart management.

Key Capabilities

• Product Discovery: Formulates queries to search product catalogs.
• Variant Selection: Identifies and selects correct product attributes like color and size.
• Cart Management: Adds products with specified quantities and variants to the cart.
• Clarification Dialogue: Engages in follow-up questions when user requests are ambiguous.
• Multi-Item Orders: Efficiently handles requests involving multiple distinct products within a single conversation.
• Tool Integration: Trained to use a suite of tools including catalog_search, catalog_get_variants, cart_add, cart_view, user_get_visit_history, and ask_user via Qwen3's native tool-calling format.

Intended Use Cases

This model is designed for integration into e-commerce platforms where it can serve as a robust shopping cart assistant. It performs optimally when:

• Connected to a real product catalog via its tool interface.
• The product catalog supports efficient text search (e.g., FAISS, Elasticsearch).
• Products include comprehensive variant information.
• The interaction involves multi-turn conversations with tool execution between turns.

Limitations

• Requires external tool implementations; the model generates tool calls but does not execute them.
• Trained exclusively on English product data.
• Variant matching accuracy is dependent on the quality of the integrated product catalog.