Chatbot for Efficient Issue Triage and Resolution

[llxia]

Motivation
The current issue triage process often relies on manual analysis and decision-making by developers, which can be time-consuming and error-prone. And the increasing volume and complexity of software issues can overwhelm development teams, leading to delays, inefficiencies, and customer dissatisfaction.

The primary objective of this proposal is to investigate if we can leverage ChatGPT to enhance the issue triage and resolution process.

Possible area of investigation
By leveraging the knowledge of ChatGPT, if we can achieve any of the following, that would be great:

• Summarize Issue
• Improved Issue Triage with prioritize queue: classify and prioritize issues based on their severity, urgency, and impact, providing developers with a clear and organized view of the task queue
• With Git commit history, extract valuable information about code changes related to a specific issue
• Use Chatbot as a virtual team member, providing real-time suggestions, code snippets, and best practices to resolve issues.
• Train and fine-tune ChatGPT: Gather historical Git issue comments, Git commit history, and other relevant test data. This dataset will be used to train and fine-tune ChatGPT to understand the project. Related: https://www.shiksha.com/online-courses/articles/how-to-fine-tune-chatgpt-for-specific-use-case/

The above ideas are just some examples. We are open to any suggestions or recommendations.

#807 is a simple example, where we provide git issue comments into ChatGPT using llama_index, then the user can ask questions. To run it, start with https://github.com/adoptium/aqa-test-tools/blob/master/Chatbot/myKeys.sh

The cost of ChatGPT can vary depending on the specific implementation and usage requirements. It is one of the risk factors that we should keep in mind.