Sweet - any tips for judging if it is working? Agent is able to respond without context so not sure how I’d be able to update my training data to improve performance