AI Infrastructure EngineersAI PractitionersAI Proving Ground DemoAI Assistants and AgentsAI Proving GroundAI & DataNVIDIA
Video
•
2:53
•
November 2, 2026
NVIDIA Switchyard Lab Demo
See how NVIDIA Switchyard routes each AI request to the best-fit model, cutting cost without giving up quality, in a hands-on lab built on the WWT AI Gateway.
Why should a one-line factual lookup cost the same as a deep architecture review? When every request goes to the same frontier model, you pay premium prices and burn premium rate limits on questions a smaller model could handle.
In this demo, Andy Moser of WWT walks through the NVIDIA Switchyard lab. Switchyard is an open library that accelerates existing model routers, so teams can build on the routing infrastructure they already use.
What you'll see:
- Passthrough routes: change one
modelfield and the same request moves from Gemma to GPT-5.5, with no new SDK. - Classifier routing: a judge model and a plain-English prompt sort requests into coding, quick, general, writing, and math or critical.
- Proof: a 17-problem coding eval, plus Prometheus metrics for routing decisions.
Next Steps
Launch the NVIDIA Switchyard lab or contact your WWT account team to discuss accelerating your existing model router.