From Google AI Studio Prototype to a Production-Ready App
Overview of Our Client
Our client required an ERP-style application for managing employees, departments, tasks, events, calendars, and internal communication. The application had initially been developed in Google AI Studio as a PWA prototype intended for use on mobile devices.
The client did not have dedicated software development expertise, so the initial functional requirements were broad and changed during development. This approach allowed the application to be built quickly, but also resulted in duplicated functionality, UX flaws, incomplete authentication, and architectural decisions that were not suitable for long-term production use.
The client needed to preserve the validated product concept and at the same time establish a foundation for further development and real-world operation.
- Industry: ERP / Human Resources & Team Management
- Application Type: Progressive Web Application (PWA)
- Timeline: 1.5 months
Challenge
The Google AI Studio prototype successfully demonstrated the core product concept, but it was not ready for production workloads. Because the original requirements were not sufficiently defined, the application contained lots of inconsistencies that would become difficult to maintain in the future.
The application used NoSQL storage even though the ERP domain depended on relationships between employees, departments, tasks, events, calendars, and communications. Authentication and RBAC also required significant improvement, while there was no proper database migration strategy, production backup process, performance baseline, monitoring, or alerting.
On top of that, the entire environment was closely tied to Google’s ecosystem, including Firebase and Cloud Run, which created both operational and cost concerns, as well as vendor lock-in.
The project therefore represented a typical vibe coding to production task: the goal was not simply to deploy the prototype, but to establish a maintainable, secure, measurable, and independently deployable production system.
Main Goals
Overall, the project focused on turning the working AI Studio prototype into an independently managed production application. Still, to make the project more manageable, we defined the following goals:
- Audit and restructure the exported AI Studio source code.
- Reconstruct functional and non-functional requirements.
- Resolve identified OWASP Top 10 security vulnerabilities.
- Correct RBAC and authentication issues.
- Introduce two-factor authentication.
- Replace the NoSQL database with PostgreSQL.
- Introduce automated database migrations with Prisma.
- Remove unnecessary dependency on Google Cloud Run and Firebase.
- Establish a reliable and cost-efficient production environment.
- Containerize the application with Docker.
- Automate application deployments.
- Establish database and file backup procedures with defined RTO and RPO.
- Introduce Cloudflare Firewall and CDN protection.
- Measure application performance and establish an SLA.
- Implement monitoring and alerting with Prometheus and Grafana.
- Introduce email and push notifications.
- Prepare the project for further spec-driven development.
- Allow the client to continue AI-assisted development independently through GitHub and tools such as Claude Code, Cursor, or OpenCode.
Project Overview
We first exported the project to GitHub and audited its architecture, authentication, authorization, database, dependencies, caching, and existing functionality.
The audit revealed duplicated functionality, RBAC and authentication problems, an unsuitable NoSQL data model, security issues, and missing production controls. Since Google AI Studio supported export to GitHub but not the reverse workflow, GitHub became the foundation for independent development and deployment.
To make the application production-ready, we replaced NoSQL with PostgreSQL and introduced Prisma for automated database migrations. We repaired authentication and RBAC, added two-factor authentication, and remediated the identified OWASP vulnerabilities. We also introduced push and email notifications to the application.
The infrastructure was moved from Firebase and Cloud Run to a dedicated VPS, with Docker, Cloudflare Firewall, CDN, scheduled database and file backups, and defined RTO/RPO requirements. This reduced infrastructure and hosting/email costs from $126 to $22 per month including VAT.
We also implemented automated CI/CD from GitHub, reducing release time to approximately 2 minutes, and introduced spec-driven development (SDD) so the client could continue development with AI coding assistants such as Claude Code, Cursor, or OpenCode.
Finally, we established performance and scalability baselines, achieving P99 latency below 300 ms at 98 RPS, and added Prometheus, Grafana, log monitoring, golden metrics, and email notifications to support ongoing production operations.
Solution
The final solution was a stable, production-ready ERP PWA. The application was restructured around clearly defined functional and non-functional requirements, with duplicated functionality removed and the user experience brought into a more consistent state.
The client also received an independent development and deployment environment that was no longer tied to Google AI Studio. This gave the team full control over the application’s source code, infrastructure, releases, and future development, including the option to use AI coding assistants within their existing GitHub workflow.
Core Application Capabilities
- Employee, department, task, event, calendar, and internal communication management
- Progressive web application optimized for mobile devices
- Corrected role-based access control (RBAC) aligned with business roles
- Email-based two-factor authentication
- Email and push notifications for different in-app events
- PostgreSQL-based relational data management
- Prisma-based automated database migrations
- Database and application file backups aligned with defined RTO and RPO requirements
- Docker-based application containerization
- Git-based CI/CD with approximately two-minute release times
- Prometheus and Grafana-based application and infrastructure monitoring
- Cloudflare Firewall and CDN-based application protection and delivery
- Independent GitHub-based development workflow without Google AI Studio dependency
- AI-assisted development with Claude Code, Cursor, and OpenCode
Technology Stack
To transform the Google AI Studio prototype into an independently operated production application, the solution used the following technology stack:
Backend / Runtime
- Docker
- Dedicated VPS
Database
- PostgreSQL
ORM & Migrations
- Prisma
Security
- Cloudflare Firewall
- RBAC
- Email-based 2FA
- Security hardening based on OWASP Top 10 findings
Edge / Delivery
- Cloudflare CDN
- Dedicated domain
CI/CD
- Automated Git-based deployment
Backups
- Scheduled database and file backups with defined RTO and RPO
Monitoring
- Prometheus
- Grafana
Notifications
- Email alerts
Development Methodology
- Spec-driven development
AI Development Tools
- Claude Code
- Cursor
- OpenCode
Results
The production-readiness project delivered measurable improvements across performance, deployment speed, security, and operating cost:
- P99 latency below 300 ms at 98 requests per second during load testing.
- Production workload aligned with approximately 200 daily active users and 500 monthly active users.
- Deployment time reduced to approximately 2 minutes through automated deployment.
- Infrastructure and hosting/email costs reduced from $126 to $22 per month including VAT, an approximately 83% reduction.
- 1 critical-severity and 4 medium-severity OWASP Top 10 security findings were identified and addressed during the audit.
- PostgreSQL and Prisma migrations replaced the original NoSQL data architecture and manual schema-management limitations.
- Docker, automated backups, Cloudflare protection, Prometheus, Grafana, and email alerting established a production operations foundation.
- The application was moved from a Google-dependent prototype environment to an independently deployable GitHub-based production workflow.
Core Team
- Frontend Developers: Adapted and restructured the PWA frontend for the production architecture.
- Backend Developers: Reworked authentication, RBAC, backend logic, PostgreSQL integration, notifications, and Prisma migrations.
- Project Manager: Coordinated requirements reconstruction, development priorities, and production-readiness activities.
- DevOps Engineer: Implemented Docker-based infrastructure, VPS deployment, Cloudflare, backups, CI/CD, monitoring, and alerting.