Re4-Scientific-Computing-Agent

Rewriting · Resolution · Review · Revision

The proposed agent incorporates a "Rewriting-Resolution-Review-Revision" logical chain via three reasoning LLMs (functioning as the Consultant, Reviewer, and Programmer) to solve complex scientific computing problems in a collaborative and interactive manner.

🚀 Key Features & Roadmap

• LangGraph Orchestra: Orchestrates multi-task, multi-step agent workflows, ensuring scalability and maintainability.

• Structured Output: Generates standardized, parsable formats (via Pydantic) to ensure consistency and reusability of intermediate results.

• Context Management: Utilizes a shared state graph to manage solution plans, code history, and review feedback across multi-turn agent interactions, improving the coherence of interactions.

• Multimodal Review: Enables the Reviewer agent to validate not just code logic but also visual outputs (plots, contours), achieving comprehensive quality control.

• Interactive Frontend: Develop a Streamlit-based UI for easier user interaction (In Progress).

📂 Project Structure

src/
├── ipynb_ver     # code in Jupyter Notebook (.ipynb) format
└── pycodes_ver   # code packaged as Python scripts (.py) format

📧 Contact

This work is presented at the ICLR 2026 Workshop on AI and Partial Differential Equations (AI&PDE) .

If you find this repository useful, please consider giving a star ⭐ and cite our conference Workshop paper.

@inproceedings{cheng2026re,
  title={Re4: Scientific Computing Agent with Rewriting, Resolution, Review and Revision},
  author={Ao Cheng and Lei Zhang and Guowei He},
  booktitle={AI&PDE: ICLR 2026 Workshop on AI and Partial Differential Equations},
  year={2026},
  url={https://openreview.net/forum?id=PMdr3fQpBE}
}