Hierarchical Repository-Level Code Summarization for Business Applications Using Local LLMs

TL;DR

This paper tackles the challenge of repository-level code summarization for business applications by introducing a two-step hierarchical pipeline that leverages local LLMs for segment-level summaries (based on AST-derived units) and aggregates these into file- and package-level summaries. It emphasizes grounding summaries in the business domain and problem context, particularly for telecom OSS/BSS, to capture intent beyond low-level implementation details. The approach demonstrates improved coverage and domain relevance across segments, files, and packages, with extensive evaluation using ground-truth data generated via GPT-4 and SME feedback. The findings suggest that domain-grounded, hierarchical summarization with carefully engineered prompts can produce meaningful, domain-aware summaries suitable for large, enterprise codebases, while maintaining privacy through on-premises LLMs. Future work points to agentic, self-reflective models, broader domain applicability, and multi-modal integration to further enhance repository comprehension.

Abstract

In large-scale software development, understanding the functionality and intent behind complex codebases is critical for effective development and maintenance. While code summarization has been widely studied, existing methods primarily focus on smaller code units, such as functions, and struggle with larger code artifacts like files and packages. Additionally, current summarization models tend to emphasize low-level implementation details, often overlooking the domain and business context that are crucial for real-world applications. This paper proposes a two-step hierarchical approach for repository-level code summarization, tailored to business applications. First, smaller code units such as functions and variables are identified using syntax analysis and summarized with local LLMs. These summaries are then aggregated to generate higher-level file and package summaries. To ensure the summaries are grounded in business context, we design custom prompts that capture the intended purpose of code artifacts based on the domain and problem context of the business application. We evaluate our approach on a business support system (BSS) for the telecommunications domain, showing that syntax analysis-based hierarchical summarization improves coverage, while business-context grounding enhances the relevance of the generated summaries.

Figures (1)

• Figure 1: Two-step hierarchical approach to repository-level code summarization tailored for business applications