Camille Roth-Svensson

Separate freshness and accuracy testing to diagnose retrieval failures.

How to coordinate retrieval timing with real-time token delivery.

AI agents need search APIs built for machines, not browsers.

Iterative verification catches errors that single-pass research agents miss.

LlamaIndex handles retrieval, LangGraph handles orchestration, each needs different search APIs.

Averaging citation scores hides failures, so score each rubric separately.