Hi @jamescalam I would like to propose a healthcare-focused semantic routing example for the Semantic Router project.
Healthcare conversational AI systems frequently need to route natural-language requests across closely related workflows such as:
- Claims
- Eligibility
- Prior authorization
- Benefits
- Pharmacy
- Infusion therapy
- Appeals
These requests are often ambiguous and overlap semantically. For example, a caller saying, “My infusion medication was denied,” may be referring to a pharmacy issue, a prior authorization requirement, a claim denial, or an appeal. Traditional keyword routing often performs poorly in these scenarios.
The proposed contribution would demonstrate how Semantic Router can be applied to healthcare member and provider requests using a reusable set of route definitions, realistic de-identified utterances, ambiguous test cases, and route-threshold optimization.
Initial deliverables could include:
- Route definitions for common healthcare administrative workflows
- Representative member and provider utterances
- Positive, negative, and ambiguous routing examples
- An evaluation dataset with expected route labels
- A route-optimization example
- A fully local example using a supported local encoder
- An example mapping routes to specialized workflows or AI agents
- Documentation describing limitations and how users can extend the taxonomy
The contribution would not include clinical diagnosis or treatment recommendations. It would focus on administrative and operational healthcare routing.
I would appreciate maintainer feedback on:
- Whether a healthcare routing example fits the project roadmap
- Whether it should be delivered as a notebook, documentation example, or reusable example module
- The preferred encoder and evaluation approach
- Whether the initial scope should remain limited to a smaller set of routes
I would be happy to implement the example, tests, documentation, and evaluation assets.
Hi @jamescalam I would like to propose a healthcare-focused semantic routing example for the Semantic Router project.
Healthcare conversational AI systems frequently need to route natural-language requests across closely related workflows such as:
These requests are often ambiguous and overlap semantically. For example, a caller saying, “My infusion medication was denied,” may be referring to a pharmacy issue, a prior authorization requirement, a claim denial, or an appeal. Traditional keyword routing often performs poorly in these scenarios.
The proposed contribution would demonstrate how Semantic Router can be applied to healthcare member and provider requests using a reusable set of route definitions, realistic de-identified utterances, ambiguous test cases, and route-threshold optimization.
Initial deliverables could include:
The contribution would not include clinical diagnosis or treatment recommendations. It would focus on administrative and operational healthcare routing.
I would appreciate maintainer feedback on:
I would be happy to implement the example, tests, documentation, and evaluation assets.