How we used DSPy to turn AI evaluations into better responses in Dash chat
Dropbox engineers used DSPy to improve the performance of their AI-powered chat agent, Dash. They created a system of evaluation judges to assess the agent's responses, which involved a wide range of factors including user intent understanding, context selection, and tool usage. By calibrating the judges against human-labeled examples, they ensured reliable evaluations and used DSPy's optimization algorithms to improve agent performance, reducing incomplete answers and token usage.
InfrastructureScale