How to create a new service¶

This guide will help you in the steps to create a new service for the Core engine.

Follow the official tutorials¶

If you have not already, we highly recommend you to follow the official tutorials to learn how to create a new service:

• Rotate an image (generic template)
• Summarize a text using an existing model (generic template)
• Detect anomalies using a model built from scratch (model from scratch template)

Choose a template¶

We have built two Python + FastAPI templates that you can use to quickly bootstrap a new service:

• Create a new service (generic) template
• Create a new service (model from scratch) template

To help you choose the right template, the following flow diagram can help you decide the right template for your use case.

flowchart TD
    which_template_to_choose["Which template\nto choose"]
    does_your_service_use_ai{"Does your service\nuse AI?"}
    how_to_you_obtain_the_model{"How do you\nobtain the model"}
    use_the_create_a_new_service_generic_template["Use the 'Create a new service (generic)' template"]
    use_the_create_a_new_service_model_from_scratch_template["Use the 'Create a new service (model from scratch)' template"]

    which_template_to_choose --> does_your_service_use_ai
    does_your_service_use_ai -->|"No"| use_the_create_a_new_service_generic_template
    does_your_service_use_ai -->|"Yes"| how_to_you_obtain_the_model
    how_to_you_obtain_the_model -->|"An existing model (binary file or existing library)"| use_the_create_a_new_service_generic_template
    how_to_you_obtain_the_model -->|"A model built from scratch (code to model)"| use_the_create_a_new_service_model_from_scratch_template

Bootstrap a new service based on the chosen template¶

Once you have chosen the right template for your use-case, you can create a new service based on the chosen template. You have three ways to bootstrap a new service based on the chosen template:

Implement your service¶

Follow the instructions mentioned in the template you just bootstrapped to implement your service with your own logic/model.

Test your service¶

We highly recommend to test your service locally by starting the Core engine, registering your service and trying your service from end-to-end.

To ensure the service can run correctly, run the following commands:

1
2
3
4
5

# Build the Docker image
docker compose build

# Start the Docker container
docker compose up

Access the service at http://localhost:8080 and validate everything works as intended.

Build, publish and deploy your service¶

Once your service is implemented and you validated it works as intended, you can publish your service to a container registry such as GitHub Container Registry with the help of Docker and Docker Compose using the How to build, publish and deploy a service guide.

If you do not want to publish your service, you can skip this section.

Update the documentation for your service¶

If you have published your service, you can update the documentation for your service by following the How to update the documentation guide.

Configure the GitHub repository for your service¶

If you store your service's code on GitHub, you can configure the GitHub repository for your service by following the How to configure a new GitHub repository guide.

Frequently asked questions¶

Which framework/language to use¶

Any framework/language can be used to create a new service compatible with the Core engine as long as it follows the specifications of the Swiss AI Center project.

The chosen framework/language must be able to receive/make requests on some HTTP endpoints.

In all Swiss AI Center's services, we have chosen to use Python with the help of FastAPI.

Where to host the service's code¶

It is not mandatory to host the service's code to use the Core engine.

If you want to host the service's code, we recommend Git with the help of GitHub.

Where to host the service's model code¶

It is not mandatory to host the service's model code to use the Core engine.

If you want to host the service's model code, we recommend Git with the help of GitHub.

Where to host the service's model data¶

It is not mandatory to host the service's model data to use the Core engine.

If you want to host the service's model data, we recommend DVC with the help of Git and GitHub.