Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ProSEA: Problem Solving via Exploration Agents

William Nguyen, Vinh Luong, Christopher Nguyen

TL;DR

ProSEA addresses the brittleness of single-agent planning in complex tasks by introducing a hierarchical, model-agnostic multi-agent framework. A Manager Agent coordinates domain-specific Expert Agents that explore solution spaces in breadth and depth, while a rich feedback mechanism captures failures, constraints, and learnings to drive adaptive replanning. The approach yields transparent exploration traces and supports human-in-the-loop collaboration, enabling robust problem solving without domain-specific training. Experiments on FinanceBench show ProSEA surpassing traditional RAG and ReAct baselines and approaching DANA's performance without requiring extensive human guidance, highlighting practical impact for real-world, adaptive AI agents.

Abstract

Large language models (LLMs) have empowered AI agents to tackle increasingly complex tasks. However, most existing agents remain limited to static planning and brittle interactions, falling short of true collaboration or adaptive reasoning. We introduce ProSEA, a modular, general-purpose multi-agent framework designed for iterative problem solving through exploration and plan evolution. ProSEA features a hierarchical architecture in which a Manager Agent orchestrates domain-specialized Expert Agents, decomposes tasks, and adaptively replans based on structured feedback from failed attempts. Unlike prior systems, ProSEA agents report not only success or failure but also detailed reasons for failure and newly discovered constraints, enabling dynamic plan refinement informed by exploratory traces. The framework operates autonomously but supports seamless integration with human collaborators when needed. Experiments on the challenging FinanceBench benchmark demonstrate that ProSEA, even without human feedback, outperforms state-of-the-art baselines and achieves robust performance across reasoning-heavy tasks. These results underscore ProSEA's potential as a foundation for more transparent, adaptive, and human-aligned AI agents.

ProSEA: Problem Solving via Exploration Agents

TL;DR

ProSEA addresses the brittleness of single-agent planning in complex tasks by introducing a hierarchical, model-agnostic multi-agent framework. A Manager Agent coordinates domain-specific Expert Agents that explore solution spaces in breadth and depth, while a rich feedback mechanism captures failures, constraints, and learnings to drive adaptive replanning. The approach yields transparent exploration traces and supports human-in-the-loop collaboration, enabling robust problem solving without domain-specific training. Experiments on FinanceBench show ProSEA surpassing traditional RAG and ReAct baselines and approaching DANA's performance without requiring extensive human guidance, highlighting practical impact for real-world, adaptive AI agents.

Abstract

Large language models (LLMs) have empowered AI agents to tackle increasingly complex tasks. However, most existing agents remain limited to static planning and brittle interactions, falling short of true collaboration or adaptive reasoning. We introduce ProSEA, a modular, general-purpose multi-agent framework designed for iterative problem solving through exploration and plan evolution. ProSEA features a hierarchical architecture in which a Manager Agent orchestrates domain-specialized Expert Agents, decomposes tasks, and adaptively replans based on structured feedback from failed attempts. Unlike prior systems, ProSEA agents report not only success or failure but also detailed reasons for failure and newly discovered constraints, enabling dynamic plan refinement informed by exploratory traces. The framework operates autonomously but supports seamless integration with human collaborators when needed. Experiments on the challenging FinanceBench benchmark demonstrate that ProSEA, even without human feedback, outperforms state-of-the-art baselines and achieves robust performance across reasoning-heavy tasks. These results underscore ProSEA's potential as a foundation for more transparent, adaptive, and human-aligned AI agents.

Paper Structure

This paper contains 12 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Limitations of Current AI Agents
  4. Human-AI Collaboration and Interaction Paradigms
  5. Iterative Planning and Multi-Agent Architectures
  6. Interactive and Iterative Problem Solving Agents
  7. Overall Architecture
  8. Expert Agents and the Exploration
  9. Iterative Exploration Through Adaptive Planning
  10. Human-in-the-Loop Exploration
  11. Experiments
  12. Conclusion

Figures (1)

  • Figure 1: ProSEA's pipeline: The Manager Agent receives the initial question and forwards it to the Problem Analyzer for preliminary analysis. The analyzed problem is then passed to the Planner, which generates a multi-step solution plan. Each step is assigned to a domain-specific Expert Agent responsible for execution through reasoning, tool usage, and user interaction. Results from each step are sent back to the Manager Agent, which evaluates the need for plan adjustment. Upon successful completion of all steps, the Manager Agent synthesizes the final answer based on the step outputs.