Voice & Chat with Natural Language Processing

Innovation Lab

Voice & Chat with Natural Language Processing

At Valence, we help enterprise customers understand and apply next-generation technologies in a smart and innovative way to advance business goals. We often experiment with these technologies within our own business, testing and tweaking how we design, develop, and deploy innovation. It often begins by identifying a simple challenge.

Challenge

After conducting research and interviews with our clients in healthcare, we identified an opportunity to address an operational bottleneck resulting from patient confusion around treatment and appointment information.

In healthcare settings, patients are often responsible for managing their own care plan. This can include being accountable for understanding their health or diagnosis, navigating healthcare systems to identify specialists, and scheduling their own care. Because of the variables and their novice understanding of the healthcare system, time is lost to scheduling errors and misunderstandings about process and providers.

Solution

Our innovation lab had previously experimented with voice and chat frameworks for other client projects. We decided to use those frameworks as the foundation for a voice and chat enabled personal assistant that could use artificial intelligence to support patients as they navigate their care.

The goals were to provide patients with a personalized experience, providing assistance specific to their care throughout the full lifecycle of an appointment. Features included driving directions, reminders, custom Q&A, and follow–up information.

We extended our voice and chat framework with a domain-specific natural language processor and symptom diagnosis engine.

Results

The proof of concept proved effective in supporting patients quickly. The framework was tested to carry out diagnoses on 520 unique patients. Based on keyboard inputs, the engine took an average of 41 seconds to give an accurate result. Voice inputs also resulted in an impressive 53-seconds average time to a correct answer.