The “Ticket-to-Code Problem” is a major software development obstacle that many tech-driven businesses with in-house software development teams must overcome. This problem occurs when needs or issues are reported and recorded as tickets that must be converted into workable code modifications. Regnology is an example of this, having pioneered global technology in regulatory reporting solutions.
Understanding the issue, examining the user story, identifying the pertinent code sections that require modification, and putting the required code changes into practice are all steps in the ticket to code process. This process can be laborious and prone to mistakes, especially in big companies with intricate codebases. McKinsey reports that ineffective software development procedures can result in a major reduction in productivity, with some research indicating that ineffective workflows might waste as much as 50% of a developer’s time.
Regnology created the Ticket-to-Code Writer, an inventive solution, to address this difficulty. By automatically converting bug reports into usable code, this technology greatly expedites the software development process. Acknowledging the possibility to improve this solution even more, Regnology collaborated with Google, utilizing both Vertex AI and their AI know-how. Through the integration of cutting-edge AI technology, the tool’s capabilities were expanded and refined, making it an even more valuable resource for developers and saving them 60% of their time.
Using Vertex AI to Address the Ticket-to-Code Problem
An AI-powered tool called Ticket-to-Code Writer is in use today and was created to speed up the process of turning bug reports into implemented code. This solution is a component of Regnology’s larger plan to improve the efficiency and precision of its software development processes by incorporating cutting-edge AI and cloud technologies into its products.
A Comprehensive Examination of the Ticket-to-Code Writer as the Technical Solution
A variety of AI agents, each with a specific area of expertise in the ticket-to-code development process, are integrated into the architecture of the Ticket-to-Code Writer. Using similarity search and other techniques, this system minimizes dangers such as hallucinations that might arise with large language models (LLMs) and ensures correctness. This is how it operates:
- Report Parsing & Summarization [1]: To determine the required code modifications, the method begins with parsing the user story or issue report and extracting important information, such as keywords and core needs.
- Code Search & Localization [3]: This module is primarily responsible for locating specific code parts that require editing. It conducts a vector similarity search using Google’s Vertex AI search, utilizing Approximate Nearest Neighbor (ANN) algorithms to comb through maybe hundreds of files. In order to maintain the system updated with the most recent code changes, this module retrieves the most recent modifications from the repository. These changes are then chunked, embedded, and kept in GCP buckets.
- Modification Plan Generation [4]: The Modification Plan Generator, which is fueled by the Gemini 1.5 Pro LLM, takes over after the pertinent files have been located. This AI agent creates a thorough technical plan by dividing the necessary modifications into doable tasks. It lists all the code that has to be added, changed, or removed along with explanations and sample code. It also determines whether or not certain files that the Code Analysis Module retrieves need to be modified. This stage allows users to interact with the process by asking improvements in plain text, making it even easier for people who are not as familiar with technical terms to understand.
- Automated Code Generation & Pull Request Creation [5-8]: This module uses the Gemini 1.5 Pro LLM to automatically generate the required code changes and submit a pull request after the plan is approved.
Managing less popular programming languages, which the LLM might not have been initially trained to handle well, presents one difficulty during this phase. Dynamic Few-Shot Prompt Composition [5] was developed in order to remedy this. Throughout this procedure, samples of prior tickets and the accompanying code implementations are provided. After storing these samples in GCP buckets, Vertex AI similarity search is once more used to find related, similar tickets. From there, the prompt for the new desired task is created. By using historical examples to guide its output, this customized method helps the LLM to produce accurate code even in programming languages it is not familiar with.
After every module, the solution includes a post-processing and validation step to make sure the output satisfies the necessary format and quality criteria. Using the Vertex AI Gen AI Evaluation Service, this procedure methodically assesses and chooses the best answers while supplying quality metrics and justifications. This architecture’s ability to benchmark and test each module independently makes it possible to identify bottlenecks in the overall accuracy and performance, which is one of its strengths.
Impact as a Whole and Next Actions
The Ticket-to-Code Writer provides a number of noteworthy advantages:
- Enhanced Efficiency: The tool saves around 60% of the time needed to fix defects and add new features by automating the transfer from user stories to code modifications. This is especially important for teams working on large projects with complex codebases.
- Improved Onboarding for New Developers: By offering an organized method for comprehending and navigating intricate codebases, the tool dramatically lowers the learning curve for new team members.
- Improved Accuracy and Consistency: The AI-driven process ensures code changes match user stories, reducing errors and improving software quality.
- Improved Collaboration and Communication: Clear documentation and an automated workflow help developers, business analysts, and other stakeholders communicate and meet project goals.
Regnology and Google are looking into adding more features and improvements to the Ticket-to-Code Writer as they work to improve and broaden its capabilities. The intention is to allow people to communicate with their code, allowing for automated code reviews, the application of code changes via tickets, and possibly even the creation of new files as needed.
Regnology intends to continue working on producing documentation from code and creating a code explainer that can respond to queries using code, e.g., explaining the creation of a certain field value in a table. The goal of this strategy is to produce a system that is even more precise, efficient, and user-friendly and that can handle the constantly changing issues associated with software development.
