MOSAIC: Multi-agent Orchestration for Task-Intelligent Scientific Coding

2025-10-09

AI Summary

To address challenges in complex scientific coding—including multi-step reasoning difficulty, insufficient domain-knowledge integration, severe hallucination in large language models (LLMs), and lack of I/O test cases for iterative refinement—this paper proposes a training-free multi-agent LLM framework grounded in a student-teacher paradigm. The framework integrates self-reflection, stepwise problem decomposition, code generation, and debugging, augmented by a Consolidated Context Window mechanism to ensure contextual coherence across long-horizon subtasks and enable domain-adaptive reasoning. Through multi-agent collaboration and reflection-driven error correction, the approach substantially mitigates hallucination while enhancing interpretability and robustness. Evaluated on multiple scientific coding benchmarks, our method achieves statistically significant improvements over existing state-of-the-art approaches in both accuracy and generalization—particularly excelling on tasks demanding deep domain expertise and intricate multi-step logical reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Multiagent Learning
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

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 training-free framework with specially designed agents to self-reflect, create the rationale, code, and debug within a student-teacher paradigm to address the challenges of scientific code generation. This design facilitates stepwise problem decomposition, targeted error correction, and, when combined with our Consolidated Context Window (CCW), mitigates LLM hallucinations when solving complex scientific tasks involving chained subproblems. We evaluate MOSAIC on scientific coding benchmarks and demonstrate that our specialized agentic framework outperforms existing approaches in terms of accuracy, robustness, and interpretability.

Problem

• Solving rigorous scientific coding tasks with domain knowledge
• Addressing interconnected subproblems requiring stepwise decomposition
• Mitigating LLM hallucinations in complex scientific workflows

Innovation

• Multi-agent LLM framework for scientific coding tasks
• Training-free student-teacher paradigm with self-reflection
• Consolidated Context Window mitigates hallucinations in subproblems