-
Feature
-
Resolution: Unresolved
-
Normal
-
None
-
None
-
False
-
-
False
-
Not Selected
-
0% To Do, 0% In Progress, 100% Done
Feature Overview
Different use cases benefit from different RAG pipelines. Until now, RAG pipelines have required a fixed definition from developers and are not reusable.
This functionality aims to create a standardized RAG pipeline schema, inspired by Ilya Kolchinsky's demo. This JSON schema allows the representation of RAG steps as modular and reusable elements, and the UI experience will make it possible for general users to benefit from the functionality.
This approach will enable RHEL AI to:
- Define custom RAG pipelines optimized for different use cases
- (in the future) enable users to define on UI or onboard custom RAG pipelines
- This schema will allow greater flexibility and extensibility for RAG pipelines when using InstructLab on Red Hat OpenShift AI (RHOAI).
Goals
- The primary user type for this feature is users and field teams looking to deliver a POC or Pilot
- This feature will differentiate Red Hat AI from traditional RAG pipelines by providing a standardized schema format.
- The expected observable functionality is the ability to create, manage, and use custom RAG pipelines in both RHEL AI and InstructLab on RHOAI.
Requirements
- The RAG pipeline schema should be portable, versioned, and reusable for RHEL AI and InstructLab on RHOAI.
- The schema should support custom RAG pipelines with clear input and output formats.
- The schema should be easily understandable and modifiable.
Background
The RAG pipeline schema is a crucial differentiator for creating, managing, and generating text based on retrieval and generation techniques. A standardized schema ensures consistency and interoperability between different Red Hat AI systems and platforms.
Done
- The RAG pipeline schema format has been developed.
- The schema is portable and can be used across Red Hat AI products
- The schema supports custom RAG pipelines
Questions to Answer
- How will the RAG pipeline schema as an artifact be shared across platforms? S3? OCI?
- Are there any performance or scalability considerations for the RAG pipeline schema?
- How will a user interface for creating and managing custom RAG pipelines be integrated with the RHEL AI UI?
Out of Scope
- The implementation details of the RAG pipeline schema, such as the retrieval and generation techniques.
Customer Considerations
- The RAG pipeline schema should be easy to understand and modify by technical product managers and AI engineers.
- The schema should support a wide range of use cases and text generation scenarios.
- The schema should be compatible with the existing AI systems and platforms used by the customer.