Introduction
AWS DNA(AWS Digital Native Architects) is a training and networking program by AWS Korea, designed to provide a systematic experience in solving business challenges using AWS services.- Each cohort carefully selects participating companies for a rigorous program consisting of 4 weeks of education, 2 weeks of Pre-POC MVP development, and a final presentation.
- Starting with the first cohort in 2020, I participated in the 6th cohort focused on Gen AI as a representative of
Gentle Monster, and our team won theCustomer Obsessionaward at the final presentation.
Motivation for Participation
- As a new member of Gentle Monster's Technology Strategy Team this year, I was working on Gen AI-related projects when I learned about the AWS DNA program during an AWS Gen AI Immersion Day session.
- Our company's AWS Solutions Architect, Seungwon Choi, introduced the program, and after expressing interest, our entire team was selected to participate.
Schedule
- The AWS DNA 6th cohort followed this schedule:
- 2024-05-29: Understanding Gen AI concepts and project conceptualization
- 2024-06-05: Gen AI theory and practical sessions based on AWS
- 2024-06-12: Bedrock with RAG techniques, use cases, Knowledge Base introduction, and Bedrock with Agent utilization
- 2024-06-19: Bedrock Advanced architecture, Bedrock Production application tips: Bedrock Security, Cost Optimization
- 2024-06-26: Team Pre-PoC development and office hours
- 2024-07-03: Team result presentations and awards ceremony
Participating Companies
- Team 1: Kakao-A (Infrastructure Technology Organization)
- Team 2: Kakao-B (AI Platform Team)
- Team 3: Ground X
- Team 4: Woowa Brothers-A (Recommendation Service Team)
- Team 5: Woowa Brothers-B (Delivery Time Prediction Service Team)
- Team 6: Interpark Triple
- Team 7: Great Imagination
- Team 8: Midas IT
- Team 9: I'mweb
- Team 10: Kurly
- Team 11: Refine
- Team 12: Gentle Monster
- Team 13: Hyundai Card
Our Project
- As a global high-end fashion brand, Gentle Monster is renowned for its unique offline customer experiences. We identified a need to provide an equally exceptional online experience.
- We developed a Gen AI-based product recommendation search engine and chatbot to demonstrate this capability.
- Using Amazon's Working Backwards approach, we defined a virtual loyal Gentle Monster customer and started with a press release to validate our project idea.
- We created a service that recommends optimal products based on customers' natural language queries, utilizing LLM prompt engineering and RAG techniques.
- We leveraged Amazon Bedrock Claude 3.5 Sonnet, Amazon Bedrock TitanText Embedding v2, Amazon SageMaker, and Amazon OpenSearch.
- Through this 6-week process, we developed a service that allows customers to receive quick recommendations through natural language and image queries, showcasing the power of Gen AI integration.
Final Presentation Photos
Reflections
- Winning the Customer Obsession award was a proud and joyful moment, especially invigorating for a 15-year veteran engineer like myself.
- Despite initial reservations about the 6-week commitment, the program provided a compressed, high-level Gen AI learning experience equivalent to what might have taken 6 months otherwise.
- AWS's technical support was excellent, and the engineer-to-engineer interactions during the program were particularly valuable.
- We were the only team to use Kotlin for our Pre-POC project, aiming for production-level implementation from the start. We even made feature requests to LangChain4j for necessary functionalities.
- The program offered insights into how other prominent companies are addressing challenges with Gen AI, providing a broader perspective and networking opportunities.
- Special thanks to our Technology Strategy Team, our CTO, and the AWS Korea team, including Program Managers Youngjoo Yoon and Jihye Shin, and SA Seungwon Choi, for their support and efforts in making this an exceptional experience.