evidently

Why we included this project

Teams running ML or LLM systems in production rarely have one place to ask 'is this still behaving correctly?'. Evidently fills that gap with a Python library that computes 100+ built-in metrics across tabular data quality, distribution drift, and LLM output evaluation, then turns them into interactive reports, test suites, or a self-hosted monitoring dashboard. The modular design is the real selling point: run a one-off evaluation in a notebook, convert the same checks into pass/fail conditions for CI/CD, or stream metrics into a live UI without rearchitecting your pipeline. It handles both predictive and generative workloads, so a team juggling classic classifiers alongside RAG or chat apps gets one toolchain instead of two. The LLM evaluation side, covering descriptor-based checks, LLM-as-a-judge, and retrieval relevance, is a good fit for anyone currently eyeballing model responses manually and wanting something repeatable.