Published
Personalized Router
GNN-based personalized router that uses user features to tailor LLM selection.
Overview
The Personalized Router uses Graph Neural Networks to incorporate user features into the routing decision, delivering tailored model selection that adapts to each user's unique requirements.
How It Works
User features (historical preferences, domain expertise, quality expectations) are encoded as node attributes in a graph. A GNN processes this graph to learn personalized routing policies that generalize across users and queries.
Strategy
Uses Graph Neural Networks to incorporate user features for routing.
API Endpoint
autoroute:personalizedrouter
Use Cases
- User profiling for customized LLM selection
- Personalized LLM selection at scale
- Multi-tenant applications with diverse user needs
Best Practices
Feature Engineering
The quality of user features directly impacts routing performance. Include behavioral features (past model choices, rating history) alongside demographic features for the best personalization.
Related Models
- GMT Router — Graph-based personalization without GNNs
- MF Router — For implicit preference learning