Walking the AI Talk: Inside DocGo’s Project Mariel
At DocGo, we are constantly exploring how technology and artificial intelligence can make healthcare more accessible. That’s why we’ve created Mariel, a text-based AI agent to help patients confirm, reschedule, or cancel appointments without a live operator. Since launch, Mariel has confirmed more than 17,000 appointments and rescheduled another 10,000, saving approximately 30 percent of call center operator time.
We sat down with Hawk Newton, DocGo CTO and the engineer behind Mariel, to learn how the project came together, why it matters, and where it is headed next.
Q: What problem were you solving when you decided to create Mariel?
Hawk: At the time, the number one barrier for our Health Plan Partners business was cancellation. We were seeing very high cancellation rates, which meant our teams would arrive at a patient’s home only to find no one there.
Obviously, that’s extremely expensive given fuel, vehicle, salary, equipment and other costs. But more importantly, it doesn’t help anyone. Every no-show means care was delayed for another patient who could have benefited from it. Reducing cancellations became critical to scaling our business and helping patients.
Q: Where did the idea for Mariel come from?
Hawk: Our old appointment confirmation tool couldn’t handle simple questions. It would ask “Do you want your appointment – Y/N” and if people said “No”, it didn’t know what to do. So I started experimenting with various large language models. Gradually, in the evenings and over the weekends, I built a tool from proof of concept to a product that helps drive efficiency here at DocGo.
From there, I connected it with our proprietary software systems: Dara, our dispatch and medical transportation platform, and our DocGo CRM. These integrations enabled Mariel to reach out to patients by text, confirm or reschedule appointments, and answer common questions about DocGo and the nature of the visit. It is also HIPAA compliant – Mariel will not share personal information unless the patient confirms and verifies their identity. We trained it on the same scripts our outreach team uses. So the conversations are consistent with what patients already expect.
Q: How did you go from an idea to a live tool so quickly?
Hawk: I believe strongly in proof of concepts. My approach is to write code quickly and see what works. For Mariel, I used Anthropic’s Claude 4.6 model combined with AWS Bedrock. That gave us important safeguards. Bedrock ensures that patient data is not used for training. It also provides guardrails against prompt injection (malicious inputs that try to trick the model into revealing sensitive information or ignoring its safety rules) and other forms of misuse.
After building the prototype, I worked with our Quality Assurance team, reviewed it with our CMO, CEO, and Chief Compliance Officer, and then piloted it with a small group of patients. The results were strong. Mariel was able to carry on a real conversation, and it gave us confidence to expand.
Q: How does Mariel handle patient experience and accessibility?
Hawk: Geography and demographics matter. For decades, systemic barriers such as language have hindered Spanish, Russian and Mandarin-speaking populations from accessing quality care. Here at DocGo we’re trying to change that with our work in states like California and New York.
Mariel is multilingual and can communicate in up to 13 languages which is essential for the communities we serve. It always introduces itself the same way: ‘Hi, I am Mariel, your DocGo concierge’ based on the preferred language of the patient. However, if it reaches out in English and a patient replies in Spanish, Chinese, or Russian, Mariel will switch languages and confirm the change in both languages so there is no confusion!our engagement and gap closure performance. This transparency enables precise, tactical deployment of our services to help our health plan partners hit their quality targets and improve financial performance.
Q: What’s next for Mariel and for AI at DocGo?
Hawk: Mariel is already scheduling and confirming appointments, and we have started expanding its capabilities to book new ones directly, especially for care gap closure services (appointments that address missed or overdue preventive care like screenings, vaccinations, or chronic condition check-ins). Deploying this update will be a major step forward in helping patients stay on track with preventive care.
We constantly seek practical ways to leverage AI to help increase operational efficiency, improve quality and drive better outcomes. To that end, we’re poised to launch an agent that will manage our medical transportation pre-billing process. We’ve tested this solution extensively and found that it saves approximately 50% of the time currently spent on this function, while meaningfully improving overall the overall accuracy of our pre-billing.
We are also developing a separate agent to review and provide quality assurance for the PCR reports that our EMTs create during their shifts. Ensuring this documentation is fully and accurately completed by the end of each shift will enable us to promptly bill insurance for these trips, and save us time that is currently spent catching issues later in the process, which requires us to follow-up with EMTs after the fact to correct their documentation.
We anticipate that the combination of these two agents will help save time and improve our revenue cycle management – which is a key metric that ultimately determines whether care delivered translates into cash collected.Hawk: Mariel is already scheduling and confirming appointments, and we have started expanding its capabilities to book new ones directly, especially for care gap closure services (appointments that address missed or overdue preventive care like screenings, vaccinations, or chronic condition check-ins). Deploying this update will be a major step forward in helping patients stay on track with preventive care.
Q: What is your view on using off-the-shelf AI tools compared to building your own?
Hawk: Most organizations have access to the same foundational models such as GPT and Claude. The advantage comes from the software you build around them. Prompts themselves do not carry lasting value. The value is in how you apply the models and the systems you create.
For DocGo, building our own tools creates a competitive advantage. It helps us differentiate from competitors, increase operational efficiency, and win RFPs. Mariel demonstrates how investing in our own technology creates real, measurable value. There is a lot of hype around AI. The reality is you can build useful and secure tools quickly when you focus on solving real problems. At DocGo we like to actually walk the AI talk.
Closing Thoughts
Mariel is one of the ways DocGo is putting practical AI to work on real operational challenges. Give it a phone and a job, and Mariel gets right to it, cutting down cancellations, saving operators from endless follow-ups, and making it easier for patients to get care. And Mariel is just getting started. With future applications like care gap closure, Mariel shows how DocGo is using AI to bring care where it’s needed most. Don’t be surprised if Mariel texts you soon…
You must be logged in to post a comment.