A developer attempted to build a fully Temporal.io-driven authentication system, integrating AI fraud detection and email verification into reliable workflows. The initial goal was ambitious: every user operation from registration to login would flow through orchestrated, observable steps. While setup with Docker Compose was straightforward, sandbox restrictions quickly complicated the pure workflow model, forcing a pragmatic shift toward a hybrid architecture.

Sandbox restrictions proved the main hurdle, blocking filesystem access and certain Python functions that configuration libraries like Pydantic required. This led to RuntimeError failures during workflow validation. The solution involved creating sandbox-safe configuration classes and isolating database operations within activities. Ultimately, a hybrid approach emerged, attempting Temporal workflows first but falling back to direct execution when sandbox issues arise, ensuring reliability.

Local AI integration with Ollama exceeded expectations, providing clear fraud risk scores and explanations for user registrations. This system, now production-ready, supports JWT authentication, a React frontend, and PostgreSQL, balancing architectural ambition with practical delivery. The project demonstrates that graceful degradation and fallback mechanisms are essential for real-world systems, offering a clear path toward incremental Temporal adoption without sacrificing immediate value.