Self-Improving Evaluations for Agentic RAG
Автор: Qdrant Vector Search
Загружено: 2025-11-12
Просмотров: 654
Agentic RAG changes how AI reasons, but it also demands modern evaluation.
Link to Slides: https://docs.google.com/presentation/...
In this session, Dat Ngo of Arize AI outlines a practical approach to evals.
He traces multi-step plans with open-source tooling, surface hidden failure modes like tool misuse and hallucinated context, and quantify not just per-turn accuracy but tool-call correctness, trajectory coherence, and multi-turn consistency. T
hen he goes beyond detection to improvement loops, routing across data sources, fixing context injection, and refining eval prompts so systems get better at judging themselves over time.
Expect hands-on examples from real deployments and guidance on instrumentation, datasets, and thresholds that matter. You’ll leave with a blueprint to make agentic systems observable, accountable, and self-improving.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
