Inspiration

There are around 1.2 million people living with HIV in the United States of America. Communities of color are disproportionately affected with nearly two to three times the rate of new infections. Stigma and institutionalized racism are among some of the challenges that continue to hamper people's access to HIV prevention, testing and treatment services. This failure to provide life-saving information about PrEP is fueling the cycle of new infections.

PrEP expert and team member, Raul, is an online PrEP educator. He utilizes his knowledge and resources to help people most in need of PrEP get the info and guidance they need… but he is only one person.

Wouldn't it be great if we had an unlimited supply of Raul to spread around cyberspace? Our team set out to do just that!

Meet Kai, powered by PrEP Paired™. Kai is an online sexual health counselor that uses the power of AI. Kai has the potential to take the knowledge and approachability of the best PrEP educators to websites, messaging apps and other conversational platforms to directly reach an audience that needs and can be benefitted by PrEP.

What it does

Kai Powered by PrEP Paired™ is a platform-independent API designed to be integrated into social apps like Twitter, Tinder and Grindr. It is able to answer natural language questions like:

  • Where to get PrEP?”
  • How to pay for PrEP?”
  • How does PrEP work?”, and many more.

Once you have all the information you need, ask Kai to recommend a clinic near you for next steps!

How we built it

Our AI Bot is designed with three simple goals in mind:

  • Simplicity
  • Responsiveness
  • Adaptability

Kai is as easy to use as any modern-day messaging app. The technology stack the PrEP Paired™ service is built on includes:

  • IBM Watson (Natural Language Processing Service Tier)
  • Ubuntu 16.04/ Nginx (OS Platform)
  • Node.js/Express (App Tier)
  • Mongo DB/Mongoose (Data Tier)
  • 3rd-party API integration: Aids.gov, Google Maps, Twitter, ArcGIS

The combination of these technologies enables our users to interact with Kai much like they would one of their online buddies. Just type in a question or statement and our software does the work of interpreting what you’re asking and then providing you with the best possible response.

Since our challenge set for the H4E Hackathon is Sexual Health Awareness and Access to PrEP, we have made a focused effort on equipping our app’s knowledge base to be well-versed in PrEP. Our service tier has been integrated into reputable 3rd-party APIs such as www.aids.gov and Google Maps, in order to further enrich the experience we can offer our users.

An equally important goal we set for ourselves was to develop our service in a way that keeps it reasonably “platform agnostic” - we can support the ability to interact with a variety of different endpoints that may want to integrate with Kai and make use of its features, including:

  • SMS Text Messaging apps
  • Messaging apps such as Twitter, Google Chat or Facebook Messenger
  • Dating apps such as Grindr
  • Traditional Web-based apps/sites
  • Native apps on a variety of mobile devices

Challenges we ran into

  • Learning the best practices involved in developing an effective grammar for understanding the “language of PrEP conversations”

  • Developing a natural and responsive flow to our AI Bot’s dialogue flow. Our goal in this was to create a conversation experience that feels fluid and casual, while also engaging our users with appropriate requests for input and feedback, in order to respond to their questions in the most effective way possible

  • Test! Test! Test! Working with a natural language processing engine is fun and challenging. It’s also a very real-world experience of “digital” meets “analog”, as you witness first hand the challenges these algorithms run into trying to flexibly accommodate the many different ways we can express the same question or thought.

Accomplishments that we're proud of

  • Successful Implementation of IBM Watson’s Conversation Service API (http://www.ibm.com/watson/developercloud/conversation.html), providing Kai with the foundation for its Natural Langauge processing capabilities

  • Development of a proprietary grammar (Intents, Entities and Dialogue) for training Kai to understand conversations surrounding PrEP

  • Designing our API so that it is “contextually aware” of what our users are trying to learn about, and where appropriate, drawing upon resources from credible 3rd-party resources such as www.aids.gov and Google Maps, to further enhance the information we provide in our responses

  • Designing our product as a scalable API that supports integration with a variety of endpoints. This has many important benefits, including a dramatic reduction in the overhead required to support other platforms and devices. It also affords our team the ability to focus on the artificial intelligence that sits at the center of KAI, and this we believe, is where the true value of product is found. Enhancing KAI to be the best-possible sexual health-care navigator on the market.

  • Establishing metrics to help us learn as our AI learns, to understand our user base and where we can make improvements. The data we track should help guide our outreach efforts and adjust our algorithm to better serve our target demographics.

What we learned

  • We have enough technology to live in an era without new HIV infections. We are not creating a solution to prevent HIV. We are just offering easy access where it is needed to resources that already exist.

  • Chatbots are fun. We really enjoyed the challenge it presented us with, both technically and creatively. The further we got along with our implementation, the more vested we all became in wanting others’ experience of Kai to feel natural. We found ourselves writing our scripts of conversation flows so we could help break ourselves out of “robotic” conversations and really keep in touch with what “messenger-style” chat feels like.

  • There’s never enough time, so prioritize! We really appreciated that being one of the ProTips that H4E shared in their motivational emails. There’s sooooo much we still want to do and working together on this project helped remind us of the importance of keeping focused on the critical goals relative to our wish-list items.

What's next for PrEP Paired

The Kai Powered by PrEP Paired™ platform could deeply integrate with services and products like Grindr, Twitter and Tinder. Kai, intelligently deployed on messaging services like these, would enable it to target other communities with unique public health challenges.

We’re eager to get Kai “into the wild” where we can begin exercising the first version of the grammar we’ve built. This will provide us with a great opportunity to learn new ways for improving Kai's adaptability and to be increasingly responsive to a diverse population of users.

We’ve also identified some near-term opportunities to deploy Kai’s use in partnership with several local health-service partners, including both municipal and not-for-profit agencies. This is especially exciting to us, because it directly ties with our goals of:

  1. improving visibility and access to PrEP,
  2. providing access an excellent online knowledgebase and,
  3. being a collaborative partner in reducing stigma around PrEP by supporting a much-needed education campaign.

So with all of this in mind, it’s full-steam ahead for Team PreP Paired!

Share this project:
×

Updates