Donut Kill My Vibe: Enriching RAG with Apache Iceberg™ for Sweet, Structured Retrieval
Open Data Lake, RAG
Unstructured text powers most Retrieval-Augmented Generation (RAG) systems. But when users want something specific, structure matters. See how Apache Iceberg™ can add high-performance, queryable structure to a RAG stack, enabling more accurate, personalized, and grounded responses.
Enter: DonutBot, an AI assistant that maps free-form donut orders—“something sweet and seasonal with sprinkles, <400 calories”—to real items. DonutBot combines semantic vector search over product descriptions with structured filters from Iceberg tables that serve as the donut metadata source of truth. This powers hybrid search that blends fuzzy LLM reasoning with precise constraints. We then use TruLens to evaluate the system, showing gains in groundedness and context relevance, yielding fewer hallucinations and better, data-backed recommendations.
Whether you're building AI for snacks or enterprise search, this talk shows how Iceberg brings structured intelligence to your RAG stack—one donut at a time.
Vino Duraisamy is a builder, strategist, and expert translator of complex technology. With a decade of experience architecting scalable data and AI solutions for industry leaders in retail, consumer tech, legal, and big tech, she possesses a rare, end-to-end perspective on what it truly takes to bring high-impact data products from concept to production. In her current role as a Developer Advocate at Snowflake, she is a key voice in shaping the go-to-market strategy for Snowflake's Data Engineering and AI products.
Josh Reini is a developer advocate at Snowflake for AI and open source. He is also a maintainer of open-source TruLens, a library to systematically track and evaluate LLM based applications. Josh has delivered tech talks, workshops and courses to thousands of developers at events like AI DevWorld.