Specialized models must handle different parts of a task. What do you build?
Why C is the answer
A model coordination system that routes each subtask
A model coordination system routes each subtask to the specialized model that suits it.
In detail
- A coordination system routes each subtask to the right model.
- Route only when a single model falls short.
Use case
An illustrative scenario, not a real incident.
A document pipeline sends images to a vision model and text to a language model, with one coordinator.
See it in different ways
flowchart LR A["Application"] --> B["Model abstraction layer"] C["Configuration"] --> B B --> D["Model A"] B --> E["Model B"]
sequenceDiagram participant S as Service participant L as Abstraction layer participant C as Configuration participant M as Model S->>L: Request L->>C: Which model? C-->>L: Model ID L->>M: Call M-->>S: Response
sequenceDiagram participant P0 as Coordinator participant P1 as Vision model participant P2 as Text model P0->>P1: Image part P0->>P2: Text part P2-->>P0: Result
Key point: Coordinate specialized models.
Why the other options are wrong
- A. A rule that only one model can ever be used in the whole application
- That rule rules out the requirement.
- B. A second AWS account for each subtask, with no connection between them
- Separate accounts add overhead and no routing.
- D. One huge prompt sent to a single model for every kind of subtask
- One prompt cannot use specialized models.