Guardado en:
Detalles Bibliográficos
Autores principales: Li, Rongjun, Zhou, Ziyu, Wu, Yihang
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2605.22057
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918516184055808
author Li, Rongjun
Zhou, Ziyu
Wu, Yihang
author_facet Li, Rongjun
Zhou, Ziyu
Wu, Yihang
contents Enterprise routers assign queries to expert agents, yet deployed profiles stay static while agents evolve (prompts, tools, models), and developers rarely keep descriptions or exemplars current. We present FlyRoute, a self-evolving profiling framework that grows capability evidence from real traffic: dispatch candidates, quality-gate successful pairs into each agent's success store, periodically distill evidence into learned capability descriptions, and inject those descriptions together with BM25-retrieved successes into an LLM router. To make this flywheel data-efficient, FlyRoute introduces a targeted exploration policy that combines profile uncertainty, BM25 relevance, and lexical novelty, prioritizing under-profiled agents only for plausible queries and avoiding redundant evidence collection. In experiments on our proprietary enterprise developer-support dataset of real routed queries, FlyRoute improves a same-backbone zero-shot LLM router from 72.57% to 78.04% with only five seed queries per agent, showing that profile retrieval already strengthens cold-start routing. After streaming 7,211 labeled training queries through the flywheel, accuracy rises to 89.83% (+17.26pp over zero-shot; +11.79pp over cold start), with consistent gains across four expert domains under standard routing accuracy on single-gold test queries.
format Preprint
id arxiv_https___arxiv_org_abs_2605_22057
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FlyRoute: Self-Evolving Agent Profiling via Data Flywheel for Adaptive Task Routing
Li, Rongjun
Zhou, Ziyu
Wu, Yihang
Computation and Language
Enterprise routers assign queries to expert agents, yet deployed profiles stay static while agents evolve (prompts, tools, models), and developers rarely keep descriptions or exemplars current. We present FlyRoute, a self-evolving profiling framework that grows capability evidence from real traffic: dispatch candidates, quality-gate successful pairs into each agent's success store, periodically distill evidence into learned capability descriptions, and inject those descriptions together with BM25-retrieved successes into an LLM router. To make this flywheel data-efficient, FlyRoute introduces a targeted exploration policy that combines profile uncertainty, BM25 relevance, and lexical novelty, prioritizing under-profiled agents only for plausible queries and avoiding redundant evidence collection. In experiments on our proprietary enterprise developer-support dataset of real routed queries, FlyRoute improves a same-backbone zero-shot LLM router from 72.57% to 78.04% with only five seed queries per agent, showing that profile retrieval already strengthens cold-start routing. After streaming 7,211 labeled training queries through the flywheel, accuracy rises to 89.83% (+17.26pp over zero-shot; +11.79pp over cold start), with consistent gains across four expert domains under standard routing accuracy on single-gold test queries.
title FlyRoute: Self-Evolving Agent Profiling via Data Flywheel for Adaptive Task Routing
topic Computation and Language
url https://arxiv.org/abs/2605.22057