Carnot: Interpretable, Interactive, and Optimized Execution of Deep Analysis Queries

arXiv:2608.09532v1 Announce Sort: cross
Summary: Enterprises more and more search to question information lakes utilizing pure language by way of AI-driven instruments like semantic operators or deep analysis brokers. Nevertheless, the latter operates as an opaque black field, hiding its intermediate reasoning and information retrieval steps, and failing to reveal controls for managing API prices and execution latency. In the meantime, the previous may be prohibitively costly for enterprise-scale information lakes. Consequently, analysts utilizing these programs lack the company to intercept hallucinated premises, confirm intermediate outcomes, or appropriate the system’s trajectory. We current Carnot, an interactive execution engine for AI-driven analytics. Carnot compiles pure language requests into bodily execution graphs and surfaces them via an interactive pocket book interface. Fairly than ready blindly for a closing output, customers can critique the plan, incrementally execute operators, examine intermediate information, or immediately edit the underlying code or semantic operator directions. Carnot’s question optimizer will optimize the question with respect to price or latency constraints offered by the person. Our demo will showcase how Carnot helps customers obtain environment friendly and verifiable insights on workloads motivated by actual enterprise use circumstances.



Source link