Customizing an LLM for Enterprise Software Engineering
The paper discusses the customization of a large language model (LLM) for enterprise software engineering, specifically focusing on Google's internal ecosystem. It details the development of Gemini for Google, which significantly improved software engineering metrics in a large-scale study. The authors provide a comprehensive methodology for adapting models to leverage internal engineering data effectively.
- ▪Gemini for Google was developed to optimize Google's internal software engineering processes.
- ▪The model outperformed baselines by reducing the mean number of iterations per turn by 23% and increasing code survival rates by about 17%.
- ▪The paper outlines a blueprint for enterprise model adaptation, including data extraction, preparation strategies, and model tuning.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Software Engineering arXiv:2605.16517 (cs) [Submitted on 15 May 2026] Title:Customizing an LLM for Enterprise Software Engineering Authors:Aditya Kini, Satish Chandra, Milad Hashemi, Saksham Thakur, Aditya Pandey, Vincent Nguyen, Marc Brockschmidt, Franjo Ivančić, Danny Tarlow, Parthasarathy Ranganathan, Petros Maniatis, Ahmed Omran, Zaheer Abbas, Anita Gergely, Martin Sevenich, Gufeng Zhang, Amy Hua, Alexander Frömmgen Ranganathan View a PDF of the paper titled Customizing an LLM for Enterprise Software Engineering, by Aditya Kini and 17 other authors View PDF HTML (experimental) Abstract:Enterprise software development is a continuous evolutionary process, characterized by incremental additions, architectural revisions, production deployments and rigorous maintenance.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.
Discussion
0 commentsMore from arXiv.org
