OnboardAssist: ML-Driven New Developer Onboarding in Commercial Software

Abstract

"Onboarding new members into legacy or large-scale software applications is often a very challenging process. New members need to adapt to the application in a short span of time. One of the major problems they face during the onboarding process is having difficulties to find project related information. Most of the time these newcomers depend on people who are currently working on these projects. This will impact the productivity of both parties. To tackle this problem, this research proposes a Machine Learning (ML) driven approach leveraging Natural Language Processing (NLP) and knowledge modeling techniques. By aggregating project information from various sources such as code repositories, pull requests, and communication channels, the methodology aims to provide context-aware guidance to new developers. Through the integration of Bidirectional Encoder Representations from Transformers (BERT) model, the approach seeks to optimize the onboarding process by improving their adaptation time and the overall productivity of development teams. Preliminary results of the prototype implementation demonstrate promising outcomes. Leveraging a dataset focused on information within agile open-source software projects, the proposed approach showcases effectiveness in providing relevant information to new developers. Initial quantitative assessments reveal an improvement in onboarding efficiency, with a reduction in the time required for new members to access critical project information. "