One of the significant apprehension for the customer was the ability to achieve desired accuracy level on the wake word utterance. As part of the implementation, we had trained the ML models on an extensive corpus of user utterances, including complex medical vocabulary, combining with retraining workflows for continuous maturity of the ML models. This helped us manage the challenges around accuracy. Our solution was built on top of a vendor Language Understanding (LU) solution, that helped us with extremely low error rates.
Available as a Search widget, the Search could be easily integrated within any other web application or collaboration platforms like Microsoft Teams, as a BOT. Besides providing access to clinical data, the Search widget provides deep linking to any enterprise applications and other BI reports, to make the data insights actionable and meaningful.
The application comes with a robust security infrastructure that helps the enterprise manage access at different levels of organization structure and users. Access profiles control what data domains the users can access and what actions they can perform on the dataset. In addition all the PII (Personally Identifiable Information) data is obfuscated by default.
With all the voice data processed on the device and behind the firewall, the application is intrinsically HIPAA and GDPR complaint.