Bridging the Prototype-Production Gap: A Multi-Agent System for Notebooks Transformation

TL;DR

The paper tackles the notebook-to-production gap in data science by introducing Codelevate, a multi-agent system (MAS) with Architect, Developer, and Structure agents that share a dependency graph to transform notebooks into modular, production-ready Python repositories. Central to the approach are Architecture Design Records (ADRs) that capture design decisions and guide refactoring under DRY and SOLID principles, integrated through a pipeline of preprocessing, dependency analysis, ADR generation, refactoring, and modularization. The authors demonstrate that autonomous code transformation preserves semantic equivalence while improving code quality metrics, enabling scalable, maintainable, and CI-friendly notebook-based workflows. This work has practical impact by enabling automated, architecture-aware notebook migrations suitable for collaborative data science and production pipelines.

Abstract

The increasing adoption of Jupyter notebooks in data science and machine learning workflows has created a gap between exploratory code development and production-ready software systems. While notebooks excel at iterative development and visualization, they often lack proper software engineering principles, making their transition to production environments challenging. This paper presents Codelevate, a novel multi-agent system that automatically transforms Jupyter notebooks into well-structured, maintainable Python code repositories. Our system employs three specialized agents - Architect, Developer, and Structure - working in concert through a shared dependency tree to ensure architectural coherence and code quality. Our experimental results validate Codelevate's capability to bridge the prototype-to-production gap through autonomous code transformation, yielding quantifiable improvements in code quality metrics while preserving computational semantics.

Figures (4)

• Figure 1: The different stages of the tool's pipeline to transform Jupyter notebooks into modular Python code base
• Figure 2: depicts the design of the Architect Agent, generating ADRs for each code cell of a Jupyter notebook and adding the ADRs to the nodes of the code dependency tree.
• Figure 3: illustrates the design of the Developer Agent, highlighting the various agent nodes and their tool capabilities. The agent iteratively refactors and validates the code before formatting it in the nodes of the dependency tree
• Figure 4: illustrates the design of the Structure Agent, showcasing the various agent nodes along with their tool capabilities to structure and group logically related refactored code into files and modules