🎥 Recorded live at the MLOps World | GenAI Summit 2025 — Austin, TX (October 9, 2025) Session Title: Hope is Not a Strategy: Retrieval Patterns for MCP Speaker: Serena Chou, Director, Product Management, Elastic Talk Track: Agents in Production Abstract: The Model Context Protocol (MCP) is emerging as a key integration layer for connecting LLMs with tools and data — but how do you make it perform reliably in production? In this talk, Serena Chou, Director of Product Management at Elastic, breaks down what happens when you move beyond “plug and pray” and start applying structured retrieval patterns to improve LLM output quality. Drawing from real experiments, Serena walks through three progressive approaches: • Naive – Plain MCP integration with unpredictable results. • Semantic – Enriching field names and metadata for modest gains. • Templated – Using structured templates and guided reasoning for consistent, high-quality responses. This practical session reveals how small structural adjustments can dramatically improve your LLM’s ability to retrieve and reason over data — without rewriting your architecture. What you’ll learn: • Why naive MCP implementations often fail in production • How semantic enrichment improves retrieval performance • How templated patterns can guide consistent, high-quality results • Emerging best practices for making MCP more powerful and reliable • Practical examples from real-world MCP experiments.