The product and the stakes

OpenEvidence gives physicians a way to look up medical evidence while they are treating patients — checking a dosage during intensive care rounds, for example, or working a case outside their specialty with no specialist nearby. Answers are drawn from peer-reviewed literature and carry citations so doctors can inspect the sources themselves.

The team behind it is small and started on Vercel with early experiments, keeping that workflow as the product grew and leaving web deployment and capacity to the platform. The demands are not those of an ordinary web app: a wrong answer can influence treatment, and downtime removes access to evidence at the moment it is needed.

"Honestly, life or death is not inaccurate," says Engineering Lead Kush Khosla, noting that the application is used in ER situations where uptime is critical. Engineering Lead Andy Yoon adds that many physicians are accustomed to outdated software, so a modern experience still has to deliver speed, stability, clarity and trustworthy information. Testing updates before production and rolling back on failure are part of that requirement.

Scale figures reported by the company: more than 20 million clinical consultations in January 2026, and more than 100 million Americans treated in 2025 by doctors using OpenEvidence.

Deploying for a team that mostly writes Python

Yoon was the only engineer on the team with a frontend background when he joined; most of the team works in Python and machine learning, and he says Vercel scaled with the team's needs given that makeup. The physician-facing application is built with the Next.js framework. Pushing to a development branch makes Vercel build and deploy a preview, which engineers use for live demos and to let colleagues try new features. Merging or pushing to the production branch triggers the production build and deployment, and a release can be reverted by rolling back to a previous deployment.

That flow supports multiple releases per day, each reaching the full physician audience:

"If I push a frontend change, I know that 650,000 doctors will see that change in about five minutes. Vercel just takes it and deploys it to everybody. That sticks with me every single day whenever I'm pushing stuff."

Kush KhoslaEngineering Lead at OpenEvidence

Traffic spikes and sustained growth without provisioning

Yoon was reading Slack when a message announced OpenEvidence had gone viral on TikTok, drawing roughly two million views in under a week. He watched traffic climb and checked the logs: response times stayed fast, error rates remained near zero, and with no extra capacity to provision he returned to his work.

"Vercel has just completely scaled with that usage. We've never had it fall over due to capacity or had to provision anything extra. Just being able to trust that it's there, to the point where we don't really even think about it, is amazing."

Andy YoonEngineering Lead at OpenEvidence

Sustained growth has been absorbed without rebuilding the web application. Khosla says he has never had to ask what five times as many doctors means for page loads or time to first paint, and that scaling happened naturally enough for the team to focus elsewhere.

Fluid compute and the compute bill

As usage grew, VP of Engineering Micah Smith tracked compute costs. After the team enabled Fluid compute for its functions on Vercel, he described the outcome alongside the wider infrastructure budget: serverless spend fell 90% with performance unchanged, and at 1000x growth Vercel accounts for less than 5% of overall infrastructure spend.

Fluid compute allows multiple requests to share a running function instance, so an instance can do other work while a request waits on another service such as a database or an AI backend. Under Active CPU pricing that waiting time carries no CPU charges, while memory remains billed for the duration of in-progress requests.

Separate projects for patients and doctors

Take Home is OpenEvidence's first patient-facing application: a doctor translates a clinical answer into plain language and shares it, and the patient logs in to review and save the evidence. Clinical-trial matching is part of the application's planned direction.

The expansion follows a pattern set by dozens of early proofs of concept, each deployed as its own Vercel project with a custom domain so stakeholders could open the applications and test the workflows themselves. Getting new projects online in minutes helped the team find product-market fit and made early enterprise partnerships easier to win. Take Home likewise has its own project and domain, so the patient experience can be updated without deploying the other applications.

"We know Next and Vercel so well, we get these independent deployments that work seamlessly and are isolated from each other. There are all these other areas we want to move into, so let's just make that a new Vercel project."

Kush KhoslaEngineering Lead at OpenEvidence