Scaling Real-Time RAG for Analytics with Qdrant

Learn how to build and scale production-grade RAG (Retrieval-Augmented Generation) systems for analytics applications in this 21-minute conference talk from Qdrant's Vector Space Day 2025. Discover GoodData's engineering approach to creating Python microservices and RAG architecture that enable natural-language analytics through real-time data processing. Explore the technical implementation of streaming semantic objects including datasets, visualizations, and dashboards into Qdrant vector database in near-real-time, followed by high-throughput similarity search and re-ranking techniques to construct precise prompts for grounded LLM responses. Examine critical aspects of ingestion patterns, schema design, and retrieval optimization strategies that effectively balance latency, recall, and cost considerations in production environments. Understand how Langfuse-based reliability testing helps identify failure modes early in the development process, along with essential metrics and tuning mechanisms that enhance production system performance. Gain insights into scaling challenges, observability practices, and implementing guardrails to ensure consistent analytical insights. Master reusable patterns for context construction, prompt packaging, and response validation that apply whether building analyst copilots or embedding AI capabilities into business intelligence workflows. See practical demonstrations of how combining real-time data processing, vector search technology, and disciplined evaluation methodologies creates robust, context-aware AI assistants for modern analytics platforms.