Replacing the Hyperice backend four weeks before launch

HyperSmart connects Hyperice’s recovery devices with guided routines, activity data and personalised recommendations. Its backend runs the accounts, integrations and recommendations behind the app.

We took the backend over four weeks before launch. Database queries were slow, the code had no coherent structure, the system crashed, and the architecture could not carry the traffic expected at launch. It also held the accounts of people already using an earlier Hyperice app.

Patching it would have kept the structure that caused those problems. We chose a full rewrite, delivered for the original launch date.

We rewrote the backend and the REST APIs behind the app and web clients in NestJS. The data model and indexes were redesigned around the queries the app actually made, with caching in Redis, and database response times fell from 1,200ms to 280ms.

Before go-live, k6 load tests modelled launch traffic against staging. The AWS infrastructure was defined in Terraform, with monitoring, alerting and runbooks in place for launch day.

The hardest part was the switch itself. Users of the earlier app had to move to the new backend without noticing. Both backends accepted the same session tokens, so anyone signed in on the old one stayed signed in on the new one. The switch took no downtime and signed nobody out.

HyperSmart launched on its original date. During the launch window the new backend carried more than 25,000 concurrent users at 99.8% uptime.

After launch we stayed on to build the product: integrations with Apple Health, Android Health, Strava, Garmin and Whoop, then AI longevity scoring and recovery recommendations delivered through notifications. The app has since grown to hundreds of thousands of monthly active users.