👋 Hello, friends! We recently sat down with Swaraj Banerjee and Larry Zhang from Sempre Health, a startup that brings behavior-based, dynamic pricing to Healthcare. They’re doing a little exciting work with machine learning and are leveraging our Expert Acceleration Program to speed up their ML roadmap.
An example of our collaboration is their latest NLP pipeline to robotically classify and respond inbound messages. Since deploying it to production, they’ve seen greater than 20% of incoming messages get robotically handled by this latest system 🤯 having a large impact on their business scalability and team workflow.
On this short video, Swaraj and Larry walk us through a few of their machine learning work and share their experience collaborating with our team via the Expert Acceleration Program. Test it out:
For those who’d prefer to speed up your machine learning roadmap with the assistance of our experts, as Swaraj and Larry did, visit hf.co/support to learn more about our Expert Acceleration Program and request a quote.
Transcription:
Introduction
My name is Swaraj. I’m the CTO and co-founder at Sempre Health. I’m Larry, I’m a machine learning engineer at Sempre Health. We’re working on medication adherence and affordability by combining SMS engagement and discounts for filling prescriptions.
How do you apply Machine Learning at Sempre Health?
Here at Sempre Health, we receive 1000’s of text messages from the patients on our platform each day. An enormous portion of those messages are messages that we will actually robotically reply to. So, for instance, if a patient messages us an easy “Thanks”, we will robotically reply with “You are welcome”. Or if a patient says “Are you able to refill my prescription?”, we now have systems in place to robotically call their pharmacy and submit a refill request on their behalf.
We’re using machine learning, specifically natural language processing (NLP), to assist discover which of those 1000’s of text messages that we see each day are ones that we will robotically handle.
What challenges were you facing before the Expert Acceleration Program?
Our rule-based system caught about 80% of our inbound text messages, but we desired to do a lot better. We knew that a statistical machine learning approach could be the one option to improve our parsing. After we looked around for what tools we could leverage, we found the language models on Hugging Face could be an incredible place to begin. Though Larry and I even have backgrounds in machine learning and NLP, we were nervous that we weren’t formulating our problem perfectly, using the perfect model or neural network architecture for our particular use case and training data.
How did you leverage the Expert Acceleration Program?
The Hugging Face team really helped us in all facets of implementing our NLP solution for this particular problem. They offer us really good advice on how one can get each representative in addition to accurate labels for our text messages. In addition they saved us countless hours of research time by pointing us immediately to the proper models and the proper methods. I can definitely say with a number of confidence that it will’ve taken us lots longer to see the outcomes that we see today without the Expert Acceleration Program.
What surprised you in regards to the Expert Acceleration Program?
We knew what we desired to get out of this system; we had this very concrete problem and we knew that if we used the Hugging Face libraries accurately, we could make an amazing impact on our product. We were pleasantly surprised that we got the assistance that we wanted. The folks that we worked with were really sharp, met us where we were, didn’t require us to do a bunch of additional work, and so it was pleasantly surprising to get exactly what we wanted out of this system.
What was the impact of collaborating with the Hugging Face team?
A very powerful thing about this collaboration was making an amazing impact on our business’s scalability and our operations team’s workflow. We launched our production NLP pipeline several weeks ago. Since then, we have consistently seen almost 20% of incoming messages get robotically handled by our latest system. These are messages that may’ve created a ticket for our patient operations team before. So we have reduced a number of low-value work from our team.
For what form of AI problems should ML teams consider the Expert Acceleration Program?
Here at Sempre Health, we’re a fairly small team and we’re just beginning to explore how we will leverage ML to higher our overall patient experience. The expertise of the Hugging Face team definitely expedited our development process for this project. So we would recommend this program to any teams which might be really trying to quickly add AI pipelines to their products without a number of the effort and development time that normally comes with machine learning development.
With the Expert Acceleration Program, we have put together a world-class team to assist customers construct higher ML solutions, faster. Our experts answer questions and find solutions as needed in your machine learning journey from research to production. Visit hf.co/support to learn more and request a quote.
