Ro Innovation, Technology & Agility

What project are you most excited to work on in 2025? 

Our team is working on an exciting LLM-based pipeline to triage the millions of incoming patient messages we receive each year. The goal is to help our incredible team of nurses, providers and patient advocates by intelligently routing messages to the person most qualified to help with that specific question or need. This will give our care teams more leverage and save them time while supporting our patients. This is compelling because it directly solves an immediate, real-world problem.

What does the roadmap for this project look like? 

Our goal is to streamline response times for patients while reducing manual triage time for our care teams. We’ll start with classifying and routing patient messages to the correct team of responders. From there, we will develop a series of LLM prompts designed to identify situations and route them with high accuracy to specialized teams. We’ll also focus on using LLM to enable context gathering, to pull in relevant details to summarize the issue concisely for the responder. This will reduce time spent searching for information and allow for quicker, more informed decision-making. 

To bring this vision to life, we will collaborate with multiple teams across the organization — from back-end and front-end engineers to designers to operations specialists to data analysts. This highly cross functional group will play key roles to ensure this new capability provides a seamless user experience. We will continuously review patient messages with human experts to refine the model’s accuracy. Coordinating these audits is a critical part of the feedback loop that will inform our iterative improvements to deliver a better experience for both patients and our care team members.

What in your past projects, education or work history best prepares you to tackle this project? What do you hope to learn from this work to apply in the future?

Over the past decade, I’ve worked on a wide range of machine learning projects, spanning natural language processing and computer vision. However, my deepest focus has always been on NLP — originally working with small language models and continuously pushing the limits of what was possible.

I identify strongly as a prompt engineer, having engaged in some form of prompt engineering as early as 2016, long before the term was widely recognized. My experience working with early deep learning architectures gave me a strong foundation in understanding model behavior, and an intuition for crafting the inputs and harnessing the outputs of models to coax out new product possibilities. These skills have only become more relevant and valuable today with the rise of powerful transformer-based architectures.

Through this project, I hope to refine my ability to bridge the gap between AI capabilities and real-world operational needs. Ultimately, this work will shape how I approach AI-human collaboration, not just in healthcare but in any domain where AI needs to function as an assistant rather than a replacement.