Yang Yang
Engineering Manager, OpenSearch AI at AWS
Yang Yang, AI R&D Manager for Amazon Web Services OpenSearch in China, is responsible for implementing machine learning capabilities in the open-source search and analytics engine OpenSearch. He also serves as maintainer for multiple plugins within OpenSearch, including ml-commons and neural. The opensearch-neural-sparse series of models he spearheaded has consistently ranked first on Hugging Face's sparse retrieval model leaderboard for several consecutive years. The team's research interests include semantic search, generative intelligent analysis, and search engine infrastructure.
Topic
The Path to Optimal Cost-Performance in Search within GenAI
As GenAI continues to advance, search engines are being assigned a new mission—comprehensive context management. OpenSearch has actively embraced this trend and explored innovations across multiple dimensions. This talk will focus on OpenSearch’s practical implementations in high cost-performance GenAI scenarios, covering sparse search, GraphRAG, and performance-oriented development based on Claude Code. Outline 1. OpenSearch — An Integrated Open-Source Search and Analytics Engine 2. Beyond RAG — GenAI Fully Embracing Agents * Context Management: Coordinated handling of memory, state, and knowledge 3. When Tokens and Skills Are No Longer Unlimited — Reassessing Cost-Performance 4. Sparse Search — When Semantics Can Be Achieved via Text Retrieval * Going further: Inference-Free * Partially interpretable, useful for effect debugging * SparseANN algorithms: balancing cost-performance and speed 5. Graph: From Pointwise Semantics to Structural Semantics * On-the-fly learning: slow! Experience mapping: approximate! Heuristic exploration: effective! * OpenSearch Graph RAG in practice 6. Using GenAI to Optimize GenAI: OpenSearch Performance Enhancements Based on Claude Code * Establishing rules for structure and consistency * Skills and SOP * Vibe and Scope 7. Q&A
