AI RESEARCH
MOSAIC: Multi-agent Orchestration for Task-Intelligent Scientific Coding
arXiv CS.CL
•
ArXi:2510.08804v2 Announce Type: replace We present MOSAIC, a multi-agent Large Language Model (LLM) framework for solving challenging scientific coding tasks. Unlike general-purpose coding, scientific workflows require algorithms that are rigorous, interconnected with deep domain knowledge, and incorporate domain-specific reasoning, as well as algorithm iteration without requiring I/O test cases. Many scientific problems also require a sequence of subproblems to be solved, leading to the final desired result. MOSAIC is designed as a.