Skip to content

Consumer AI app · Web, iOS and Android · 2026

A vibe-coded consumer AI app that slowed at 300 users, rebuilt to run calm at 1,000+

The system slowed down at around 300 active users. After the rebuild it carries 1,000+ without slowdowns.Active users handledBefore, slowing~300After, smooth1,000+

Active users the system handles

The system slowed down at around 300 active users. After the rebuild it carries 1,000+ without slowdowns.

Before, slowing
~300
After, smooth
1,000+

Challenge

The whole system slowed down at around 300 active users, its microservice split added overhead instead of capacity, and an audit found 12 vulnerabilities and data-leak points in an app holding users' personal data.

Outcome

Fewer, cleaner services, a tuned database and closed leaks. The app carries 1,000+ active users without slowdowns, the codebase is 30% smaller, hosting costs are 15% lower, and month one after go-live had no critical bugs, outages or database slowdowns.

Key results

  • 300 → 1,000+

    active users, now with no slowdowns

  • 12

    vulnerabilities and data-leak points fixed

  • -30%

    smaller overall codebase

  • -15%

    hosting costs

Client
Consumer AI app on web, iOS and Android
Engagement
3 months, 2026
Team
4 developers
Status
Rebuilt system live

The situation

A consumer AI app turns users' personal data into personalised forecasts and reports, with an AI chat that answers questions about them. It runs on web, iOS and Android and earns through subscriptions and in-app credits. The vibe-coded version proved people wanted the product. Its backend was about ten NestJS microservices on AWS, covering forecasts, data, chat, payments and notifications.

Then it hit its limits:

  • Performance: the whole system slowed down at around 300 active users, far below where the business needed to go.
  • Architecture: the vibe-coded service split added overhead instead of capacity, and the database couldn't carry the target load.
  • Security: our audit found 12 vulnerabilities and data-leak points, in an app holding users' birth data and personal questions.
  • Maintainability: the most tangled services were risky to change, so every fix slowed the team down.

What we did

  • Audit: load, security and code review across all services before changing anything. It surfaced the 12 vulnerabilities and data-leak points.
  • Rewrite where it mattered: we rewrote most of the services and merged some of them, cutting the overhead between them.
  • Database optimisation: serious work on the database so the target load became realistic, not hopeful.
  • Refactoring: a general refactor of the most tangled services, so the team can change them safely.
  • Security fixes: we closed all 12 vulnerabilities and data-leak points found in the audit.

Results

The app now handles 1,000+ active users smoothly, up from about 300 when it started slowing down. All 12 vulnerabilities and data-leak points are closed. The codebase is 30% smaller and hosting costs are 15% lower. In month one, the first month after the rebuilt system went live, there were no critical bugs, outages or database slowdowns.

Carrying 1,000+ users took fewer, cleaner services, a tuned database and closed leaks, not more of what the vibe-coded version already had.

Stack

NestJSAWSMicroservicesDatabase optimisationLLM-based AI chatWeb, iOS and Android

Tell us what you're building, or what's breaking

A free 30 to 60 minute call with our COO and a senior engineer. You leave with a clear next step, whether or not it's us.