Clinical-Grade Uptime at Startup Velocity

When OpenEvidence went viral on TikTok, the moment passed without drama. Lead Frontend Engineer Andy Yoon watched the traffic spike—roughly two million views in under a week—and checked the metrics. Response times held. Error rates stayed near zero. There was nothing to fix.

"Vercel has just completely scaled with that usage," Yoon says. "We've never had it fall over due to capacity or had to provision anything extra."

The incident was a stress test that OpenEvidence didn't have to prepare for. For a clinical decision support platform, the stakes are unusually high: a general-purpose AI model can occasionally be wrong, but a tool guiding real medical decisions cannot tolerate that failure mode. The platform supports over 20 million clinical consultations per month and, according to the company, over 100 million Americans were treated last year by a doctor using the product.

One Frontend Engineer on a Python Team

When Yoon joined OpenEvidence roughly three years ago, he found himself in an unusual position: the only team member with a dedicated frontend background in a group dominated by Python and machine learning expertise. That constraint shaped the architecture.

The stack is hybrid. The backend runs on Python and Google Cloud Platform, handling data ingestion, model orchestration, and core business logic. The frontend, built with Next.js, lives on Vercel. Each commit triggers an automatic deploy; production pushes take about five minutes, and preview URLs are generated for every branch. For a small team supporting a platform used by a significant share of U.S. physicians, the simplicity justifies the setup.

"Given the makeup of our engineering team, Vercel has really scaled with our frontend so well," Yoon notes.

Prototyping Without Friction

OpenEvidence didn't start as the product it is today. Early on, the team iterated through dozens of proof-of-concept projects, each deployed on Vercel as a standalone app with its own custom domain. That workflow let stakeholders click through realistic interfaces long before production hardening began.

Preview deployments serve a similar purpose during feature development: shareable links for live demos, instant rollback if something goes wrong. The ability to spin up a production-lookalike environment in minutes helped the team find product-market fit and close early enterprise deals.

The 90% Cost Surprise

As OpenEvidence scaled—roughly 1000x growth since launch—VP of Engineering Micah Smith kept an eye on infrastructure spend. When Vercel introduced Fluid compute, combining on-demand execution with server-like persistence, the team decided to try it. The result was a 90% drop in serverless costs with no meaningful change in performance.

"We reduced our serverless spend by 90% while maintaining the same performance, and even as we've scaled up to 1000x growth, Vercel is less than 5% of our overall infra spend." —Micah Smith, VP Engineering

Lower latency and fewer cold starts came along as side benefits. Infrastructure became an afterthought, freeing engineering time for product work instead of capacity planning.

Reliability as the Ingredient Doctors Notice

Clinicians are accustomed to dated hospital software that is dependable, if not pleasant. OpenEvidence had to match that reliability bar while still looking modern. Yoon acknowledges the tension: "A lot of doctors and medical professionals are used to really outdated software."

The viral TikTok moment validated that balance. The platform took the traffic surge in stride, kept its uptime record intact, and preserved the trust that matters more than uptime alone. OpenEvidence now serves over 40% of U.S. physicians across more than 10,000 hospitals, yet the frontend team remains small—and the infrastructure still runs without active babysitting.