Artificial Genius Orthix version 1

Platform Configuration

The Orthix platform is delivered as a container, for deployment in ECS, requiring access to the following resources:

Bedrock Runtime (port 8080): access Orthix non-hallucinatory language models. S3 (ports 80, 443): access unstructured data stores in S3 buckets. HTTP (ports 80, 443): access unstructured data on the Web.

To support this access:

These ports should be specified in the portMappings of the ECS Task specification. In your IAM configuration, the Task/TaskExecution roles specified in the ECS Task specification will need access to the above necessary services, or to be trusted to assume the roles of principals who do. The MANAGED_INSTANCES ECS orchestration i recommended so that the set of roles that may need such IAM trust relationships is fixed. Your VPC or other network settings will also need to be configured to permit connectivity to these services. The ECS Task specification should specify the healthCheck as a /ping request on port 8080.

Model Configuration

The Orthix non-hallucinatory language models are delivered via Bedrock Custom Model Import. This is a secure service that provides inference endpoints within your AWS account, by hosting and running the Orthix models in a secure, AWS-managed escrow account.

Once the import is complete, the models will be accessible to you via model deployment ARN(s). You will need to provide these ARN(s) to the teams using the platform, which will be entered into their PRD prompts in the Agent field.

Running a Workflow

To run a workflow, the Orthix platform uses the standard container endpoint for Bedrock AgentCore agents is the REST endpoint /invocations on port 8080.

The POST input for the container endpoint is JSON with a 'prompt' field containing the PRD in Markdown format, and the output is a JSON formatted log of successes and failures of workflow items.

The results of running the workflow are written to the Output server specified in the PRD.