AI Content moderation service

This service classifies text into offensive speech categories via a REST interface, leveraging the KoalaAI/Text-Moderation model for content analysis.

See the OpenAPI documentation under /docs for details on how to use it

Monitoring

Prometheus is used to collect metrics of this service during runtime and exposed via /metrics

Requests per second

The service also tracks the incoming requests per seconds (rps). The metrics endpoint can be queried to get the current value of requests_per_second

Start the application

1. Run the server with uvicorn main:app
2. Open a browser window and naviagte to http://127.0.0.1:8000/

Prerequisites

• Install dependencies: pip install -r requirements.txt

Further improvements

To run this service in a productive environment, more work is needed:

• Hardware (GPU/TPU) should be used inference where available
• Currently the model is pulled from the web on statup -> make it available offline
• Add authentication and transport encryption, eg. through an API gateway
• Use an external monitoring system (eg. Grafana) to determine rps across several instances of this service