This article is guest authored by Justin Mardjuki, CEO, Sayvant.
Emergency rooms and urgent care facilities handle an estimated 350 million visits each year. Acute care environments are fast-paced, unpredictable, and high stakes. In these settings, providing care quickly is paramount. That’s why we created Sayvant, the leading clinical documentation AI solution for acute care. Sayvant helps clinicians generate accurate, defensible medical charts in seconds, enabling them to focus on delivering patient care instead of administrative tasks like documentation. Sayvant automates the creation of clinically accurate charts from ambient listening and clinician dictation, saving emergency medicine and urgent care clinicians more than two hours a day and improving alignment with revenue cycle management (RCM), quality, and risk teams.
We began developing Sayvant in 2024 in close collaboration with Vituity, the largest physician-owned and led multispecialty partnership with 6,000+ clinicians caring for more than 10 million patients annually. The partnership was led by Inflect Studio, Vituity’s innovation and incubation arm. The team selected Azure as our cloud partner as we began to build out the Sayvant solution. Using Azure Cosmos DB, we can power millions of discrete large language model (LLM) agentic workflows that turn encounter transcripts into robust, clinically validated medical charts.
Emergency rooms can be hectic—they need technology that can keep up
If you’ve ever been in an ER or urgent care facility, you know the environment can be high-stakes and intense. Doctors might see 40 patients in just one shift, managing several patients’ care at once. They rarely have time to retreat to a quiet space to complete documentation. Robust and accurate documentation can be challenging to create amid the chaos of an acute-care setting. As a result, poor documentation can lead to delayed patient discharge, inaccurate or incomplete billing, heightened malpractice risk, and lower adherence to quality measures.
Sayvant was built to address these challenges. Hundreds of clinicians shaped the platform’s design, providing valuable feedback as we optimized the AI-powered documentation workflow that reduces charting time, improves accuracy, and allows clinicians to focus on patient care. Since we launched a private beta in April, we’ve been privileged to see rapid adoption of Sayvant. We’ve expanded our partnership from fewer than 10 sites of care to over 50 in just six months.
With hundreds of clinicians trusting Sayvant on shifts every day, we’ve been able to meet demand, in part, thanks to Azure Cosmos DB and other Azure services. In a nutshell, Sayvant takes speech-to-text transcripts and transforms them into summaries for medical charts. We capture transcripts from conversations between patients and clinicians, as well as dictation from providers about things like imaging and lab results. Sayvant then produces charts that are tailored to the clinician’s preferences. Finally, Sayvant cross-checks the draft chart against peer-reviewed medical literature, as well as input from revenue cycles, billing, and quality teams to verify that it’s robust and defensible.
‘If Azure Cosmos DB can support ChatGPT’s scale, we knew it would be performant for Sayvant’
We knew Azure was the trusted technology partner in healthcare, and it would deliver the compliant, secure, and robust infrastructure we needed. We went all in on Azure at inception—100 percent of Sayvant runs on Azure. And the fact that Azure Cosmos DB already supports OpenAI and ChatGPT gave us confidence it would meet our needs and help us scale for the future.
Today, we deploy Sayvant with dedicated applications, servers, and Azure Cosmos DB instances at every site using infrastructure as code. That lets us meet data security and privacy requirements for our hospital customers, and guarantee high-volume performance. With Azure Cosmos DB, we can provide multi-region replication so our customers can seamlessly jump from their recovery site in a different region, and back.
We’re also able to keep chart latency low and power our multi-agent use case. By building on Azure Cosmos DB, clinicians can tap a button to generate a chart on their way out of the patient’s room, and by the time they’re back at their computer, a fully drafted chart is waiting for them. Since we launched Sayvant, clinicians have completed 15,000 shifts using the application and we’re generating thousands of charts every day, with 99 percent of charts generated in 60 seconds or less.
Powering hundreds of LLM agents
While audio transcription and summarization is table stakes, the magic of Sayvant is in its agentic infrastructure that helps clinicians generate robust, clinically-defensible charts that include medical decision-making and differential diagnosis. Hundreds of discrete LLM agents powered by Azure Cosmos DB and Azure OpenAI Service perform tasks associated with different sections of the chart. Certain agents have very simple tasks, like looking at a summary and generating an output. Others perform RAG searches and pull in peer-reviewed literature to help defend a clinician’s course of care.
Each input/output is stored in Azure Cosmos DB, and we’re processing over 2 billion LLM tokens a month. Translating that into words on a page, it’s like having 60 pages of input/output for every chart created.
Delivering last-mile personalization
The opportunities to personalize the final chart format so it matches a clinician’s preferences and voice are virtually endless. We built Sayvant around a canonical medical chart and workflow, then fine-tune Sayvant for each site so it produces charts that meet the group’s billing, quality, and risk requirements. Finally, we add last-mile personalization where Sayvant generates chart language in the voice of particular clinicians. Some doctors write in more lengthy, academic prose, while others prefer concise bullet points, and Sayvant reflects that behavior.
Optimizing for accuracy and performance
As Sayvant continues to scale, we’re always looking to optimize accuracy and speed. Our North Star metric is the accuracy of our charts, which we track through the percentage of unchanged charts that are approved by doctors. AI hallucinations in healthcare are unacceptable, and Sayvant is architected to ensure final charts reflect encounter truth.
Through Azure Cosmos DB analytics, we’re able to track metrics like latency and charts per user. We also run ongoing internal reviews comparing Sayvant-generated charts against human-generated charts. We have clinicians rate Sayvant charts and so far, we’ve found Sayvant-aided charts are more defensible in medical malpractice cases than the human-generated ones.
As Sayvant becomes an integral part of acute care across the country, we want to ensure that we continue to generate massive time savings for clinicians without sacrificing chart quality.
***
About the author
Leave a review
Tell us about your Azure Cosmos DB experience! Leave a review on PeerSpot and we’ll gift you $50. Get started here.
About Azure Cosmos DB
Azure Cosmos DB is a fully managed and serverless NoSQL and vector database for modern app development, including AI applications. With its SLA-backed speed and availability as well as instant dynamic scalability, it is ideal for real-time NoSQL and MongoDB applications that require high performance and distributed computing over massive volumes of NoSQL and vector data.
Try Azure Cosmos DB for free here. To stay in the loop on Azure Cosmos DB updates, follow us on X, YouTube, and LinkedIn.
0 comments
Be the first to start the discussion.