CASCA: A Self-Improving Multi-Model LLM Router with Quality-Guaranteed Cost Optimization

Read full text Summarise with VIRAT
Authors: Vast Intelligence Limited, CHIH CHENG CHEN
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We present Casca, a production LLM routing system that reduces enterprise AI API costs by 30–60% while maintaining quality SLA guarantees. Unlike existing routers that apply static rules, Casca implements a closed-loop Auto-Learn flywheel: real customer traffic generates ambiguous cases, human annotators label them, and the classifier automatically updates — making routing accuracy improve continuously over time. This article documents our system architecture, routing engine v1.2 (26 rules, 1,000+ labeled training samples), multimodal routing design, and the Auto-Learn pipeline, establishing a public record of our technical approach as of the publication date above.

Fields of Study & Subjects

Cloud Computing and Resource Management Artificial Intelligence Applications Software System Performance and Reliability

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
28 March 2026
Language
EN