We envision a time when consumers can sign up for Medicare, and know what they’ve signed up for and what is required of them. Everyone should be able to say:
- I’m in charge of my health care benefits.
- I have chosen the right plan for me.
- I can predict what I will be paying out of pocket, given different types of medical events.
- I know my rights and feel empowered when working with Medicare.
- I know I have complied with all rules and regulations.
- I have a community of friends and experts I can rely on.
- We'll provide transparency and clarity for the consumer.
To describe the user story, problem statement and target persona, and empathy map for our target persona, we have included three images in the image gallery: "MedicareMojo Overview - User Story", which depicts the needs of our users and how MedicareMojo addresses those needs, "Problem Statement and Target Persona", and "Empathy Map for Target Persona".
What it does
MedicareMojo helps a prospective Medicare applicant find her way through the maze of questions and decisions that must be made to initially enroll as well as to re-enroll yearly. MedicareMojo analyzes the defining Medicare documents provided by the Center for Medicare and Medicaid Services (CMS) and other authoritative sources in their electronic document repositories. Users can then conversationally interact with MedicareMojo to zero in on the enrollment questions that must be answered for their particular case.
The image entitled: "Solution - Journey Map using MedicareMojo" in the image gallery provides a visual representation of what MedicareMojo does and how it addresses the needs of our users. There are also "MedicareMojo Client Screen Shots" in the image gallery so you can see the client-side flow for the user.
How we built it
The architecture of the MedicareMojo app can be seen in the "MedicareMojo Architecture" diagram in the image gallery above.
We built the system to first ingest the authoritative CMS document corpus. We first tried using the Watson Document Conversion component and had some difficulty processing the rather raw PDFs in the corpus. We fell back to using hand construction of a JSON file to encode a number of representative issues addressed in the CMS document which we are initially considering, "Medicare and You 2016", the top-level customer targeted CMS document. This JSON file is held within the MedicareMojo Service -- our next version will use MongoDB. The JSON file is used to train an instance of the Watson Natural Language Classifier.
Questions are composed by the user on the MedicareMojo Mobile Client and transmitted to the MedicareMojo Service which uses the trained Watson NLC to return a reference to the user to the authoritative document which will answer their question.
See a sample interaction in the "MedicareMojo Mobile Client Screen Shots" in the image gallery above.
Challenges we ran into
We tried using the document conversion service, but the JSON returned wasn't useful for our application. As a work around, we parsed the document ourselves with the schema we wanted.
We joined the hackathon a couple of weeks in. This gave us little time. But we were fortunate enough to be joined by a great UX team, whose work validated our direction and quickly provided solid deliverables for us.
Accomplishments that we're proud of
We're most proud of getting the natural language classifier being useful. At first, we were unsure the classes returned back would be useful for a user to ask it a question. But after testing with different training sets, we were happy with one that yielded fairly good results.
We're also proud of the interviews we had with potential users. The things our target folks said were in line with what we were experiencing personally, and what we were seeing in the research.
What we learned
There's a lot more testing needed for training sets. We understand more deeply how to make good training sets and which ones are utter rubbish. We also learned how hard it is to do document conversion for PDFs.
What's next for MedicareMojo
We are going to continue working on this project, breaking it down into phases.
We have a lot of documentation to consume. The Social Security Administration (SSA) and Centers for Medicare & Medicaid Services (CMS) have made improvements in putting electronic versions of their content online. But consuming that content programmatically takes another level of structure. We will be creating a better way of automating document conversion with our data schema.
Also, we'll be moving on from this exploratory phase to begin to make decisions about the technology stack we use, and how we use it. For example, we want to explore other search tools, like ElasticSearch.
But folks can't wait. So, we'll be accelerating the front-end community and resources. We want our target audience to be in the mix with us, calling out the issues and designing the product they want.
For a more detailed presentation of MedicareMojo, please see the attached presentation provided in the Try it out section below using the docs.google.com link.