How would you design an inference service that meets low-latency requirements for real-time document analysis for Docusign?
