Dongwon F&B's Journey to Innovate Shopping Experiences with an AI Shopping Assistant Based on Amazon Bedrock AgentCore

Dongwon F&B carried out a project to build a next-generation AI Shopping Assistant to address the limitations of its existing health food consultation chatbot, Baro.D, which suffered from RAG-centric search limitations, non-personalized responses, and a lack of operational visibility.

Through a 6-week sprint under AWS's EBA (Experience-Based Acceleration) program, they implemented an intelligent agent that cares for the entire shopping journey, from product discovery to purchase and CS. The key goals set were achieving a TTFT (Time To First Token) of under 2.0 seconds and a RAG Precision@10 of 90% or higher.

As a core technology, they introduced Amazon OpenSearch Service to innovate search quality. They improved Korean search accuracy through the Nori Korean morphological analyzer, and applied hybrid search (RRF method) combining kNN vector search and BM25 to simultaneously implement meaning-based semantic search and precise keyword matching.

Through this project, Dongwon Mall expects the following business outcomes:

• Recommended product CTR improved from 31.5% to 35%
• CS inquiries reduced by 28%
• Providing natural language-based personalized product recommendations and real-time order/delivery/benefits inquiry features