Yes! We've been thinking about techniques like pushing the queried embeddings into a datastore to detect anomalies and track outliers. Provide some insights there.