Evaluating retrieval quality at scale
Good answers start with good retrieval. We share the evaluation harness we use to measure recall and precision across millions of documents. The post includes the metrics that…
Author
Author alpware
Smaller models, bigger wins
Bigger is not always better. We benchmarked a family of smaller models on routing tasks and found they matched larger ones at a fraction of the cost. We…
Designing guardrails that hold
Guardrails fail when they are bolted on late. This post explains how we model policy as code so the same rules apply in testing and production. We also…