Reproduce “DeepEye: A Steerable Self-driving Data Agent System” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Boyan Li, Yiran Peng, Yupeng Xie, Sirong Lu, Yizhang Zhu, Xing Mu, Xinyu Liu, Yuyu Luo

DeepEye, developed at The Hong Kong University of Science and Technology (Guangzhou), introduces a steerable, self-driving data agent system that uses a database-inspired workflow engine to autonomously execute complex, multimodal data analysis. It orchestrates analysis across heterogeneous data sources, mitigates context explosion through hierarchical reasoning, and enhances reliability and efficiency via validation and parallel task optimization.

Key insights

  • Explicit workflow orchestration through Directed Acyclic Graphs (DAGs) enhances transparency, management, and optimization of complex data analysis tasks.
  • Decomposing complex reasoning into hierarchical, context-isolated sub-agents effectively addresses LLM limitations regarding context window size and multi-step reasoning.
  • Applying database management system (DBMS) principles (compilation, validation, optimization) to LLM-driven agent execution improves reliability, efficiency, and auditability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.