AI Evaluation is often presented as a long list of independent techniques: Rule-Based Evaluation, Human Evaluation, LLM-as-a-Judge, Offline Evaluation, Online Evaluation, Component Evaluation and End-to-End Evaluation. But in practice they all complement and build on each other.
I built Wayfinder, a reference implementation where I explore these concepts progressively using the same AI application. My goal was to build an intuition for how the different evaluation techniques fit together, and how to turn them into an evaluation strategy for an AI application.
I'd be interested to hear how others are approaching evaluation in production AI applications and what you think is missing from this implementation.