AI Assistants in Healthcare: Open Source Starter Pack

Medicare Locator (alpha) is an example AI assistant to help developers in healthcare build their own.

➡ You can find the repo here

Healthcare costs are exploding all over the world. At Rasa we believe that open source technology has a massive potential to help deliver better results at lower costs. And today we're excited to make our first open contribution in healthcare.

We open sourced a starter pack for developers to learn how to automate full conversations in healthcare (Github repo here). This is intended as a learning tool and not a fully trained AI assistant. Our goal is to inspire developers all over the world to build AI assistants that help make their healthcare system more pleasant and efficient for patients, payers, providers and other stakeholders.

Use Case

Finding local healthcare in the U.S. can be cumbersome, especially if you're not exactly sure what is covered by your plan. We've created Medicare Locator to help those recipients who need the most help find nearby hospitals, nursing homes, and home health agencies that are registered with Medicare. Users can find the information they need simply by talking and asking questions.

We picked this use case to demonstrate how an AI assistant can handle a full conversation and pull in data from 3rd party APIs. Just grab the starter pack from Github and run it on your machine!

Under the hood

Medicare Locator is fully open source and consists of three major components:

Rasa : The open source framework Rasa understands the natural language input of the end user and uses machine learning to extract entities like ZIP codes city names from that text. It then uses this information to either call the Medicare API or asks the user to provide more missing information.
Medicare API: The dataset that contains the location information comes from data.medicare.gov - it contains a nationwide list of hospitals, nursing homes and home health agencies.
Chat API: Medicare Locator can connect to any conversational channel - from text message to Facebook messenger. In this example, we chose Telegram.

What's next?

The Medicare Locator is currently in alpha. We'll be collecting more training data to improve its performance and plan to add more capabilities over time, like a physician search. It's all open source - so please create an issue on Github if you have ideas for how to make the project more impactful.

This recording of our developer webinar gives you a deep dive on the medicare locator.

Do you have other ideas about how AI assistants can improve healthcare? Join the discussion in our community forum or get in touch!

Curious about other use cases in healthcare? Learn more here.