arXiv cs.AI / cs.LG / cs.CL·1d agoRECAST: Learning to Compute the Right Context through Adaptive Evidence Routing#rag#recast#qwen3.5AI research1