Published
GMT Router
BymeAI Team
Graph-based personalized router that learns user preferences for tailored LLM selection.
Overview
The GMT Router learns individual user preferences through graph-based modeling, creating personalized routing decisions that improve with each interaction.
How It Works
User interactions are modeled as a graph where nodes represent users, queries, and models. Edge weights encode satisfaction signals, preference patterns, and contextual factors. Graph traversal and propagation algorithms infer the best model for each user-query pair.
Strategy
Learns user preferences through graph-based modeling.
API Endpoint
autoroute:gmtrouter
Use Cases
- Personalized recommendations for individual users
- User-specific routing based on past behavior
- Applications with diverse user populations
Best Practices
Preference Drift
User preferences evolve over time. Implement decay weights on historical interactions so newer signals carry more weight than older ones.
Related Models
- Personalized Router — GNN-based personalization
- Graph Router — General graph-based routing without personalization