Discover how Japan's largest C2C marketplace leverages Generative AI in this 20-minute conference talk from FC Tokyo 2024. Learn from Senior ML Engineer Teo Zosa as he details Mercari's implementation of AI-powered productivity tools and features designed to enhance user experiences. Explore the journey of developing AI listing support, similar looks recommendations, and improved search relevance through fine-tuned embedding models. Gain insights into the technical aspects of model iteration, tracking, and the creation of robust embedding pipelines. Follow along as Zosa demonstrates how visual tools facilitate team collaboration and monitoring, ultimately leading to enhanced marketplace functionality. Understand the strategic approach to implementing GenAI solutions that align with business objectives while delivering tangible value to users. The presentation concludes with valuable takeaways on continuous improvement and an invitation for potential collaboration opportunities.
How Generative AI Powers the Next Generation of Mercari Marketplace
Overview
Syllabus
- Introduction by Tio discussing AI's role at Merari
- Overview of Merari's growth and marketplace features
- Formation of the dedicated AI LLM team
- Introduction of productivity tools for internal teams
- Launch of AI listing support to simplify user experience
- Explanation of the user experience for listing items using AI
- Discussion on the similar looks feature for recommendations
- Technical details about fine-tuning the embeddings model
- Importance of weights and biases for model iteration and tracking
- Next steps for building a robust embeddings pipeline
- Addressing search relevance issues with AI support
- Use of visual tools to monitor results and facilitate collaboration
- Successful iteration leading to improved relevance in search results
- Key takeaways emphasizing customer value and continuous improvement
- Acknowledgment of the teams and invitation for potential hires
Taught by
Weights & Biases
