Project Journey: Jira Automated Estimation Tool

Inspiration

The inspiration behind this project came from observing how much time Agile teams spend on manual story point estimations during sprint planning. With time-sensitive meetings and a need for precise planning, automating story points seemed like a valuable solution to improve productivity and consistency.

What We Learned

Throughout the project, we gained a deeper understanding of Jira’s API and the powerful capabilities of JQL (Jira Query Language) in searching and retrieving relevant issues. We also learned how to design effective prompts for the Atlassian Rovo AI Agent, ensuring it could calculate story point values accurately and display the results efficiently. Additionally, we optimized estimation algorithms using data-driven insights, such as historical story data, to enhance accuracy and reliability.

Building the Project

The project was structured into three main stages, as shown in the design diagram: High-Level Design, Search Design, and Calculate Story Point (SP) Design.

High-Level Design: We began by creating an Atlassian Rovo AI Agent tailored for story point estimation. The agent is designed to call our backend functionality, which handles the core actions of searching for relevant stories and calculating the story point value. This involved designing prompts specifically crafted to guide the agent in interacting with the backend system, retrieving the necessary data, and initiating the calculation process. Once the story points were calculated, the agent displayed the results in the chat window, along with a detailed breakdown of the similar stories used in the process.
Search Design: We used Jira's API and JQL (Jira Query Language) to retrieve relevant historical data. In this stage, the tool performs a parallel search process, divided into range-based batches for optimized performance. For each batch, the algorithm calculates between the current story and historical stories based on summary text similarity, filtering down to stories with a tailored percentage title similarity.
Calculate SP Design applies a weighted calculation, where each similar story contributes to the final story point estimate based on its relevance score and historical data. By assigning higher influence to more relevant stories, this approach ensures that the estimate reflects patterns from the most comparable past items, enhancing accuracy.

The structured design enabled a systematic build, making it easier to test and refine each component independently, ultimately leading to a robust and accurate estimation tool.

Challenges Faced

One of the key challenges was learning to work with Atlassian's newly released Rovo AI framework. Understanding how to create meaningful and effective prompts that align with our tool’s functionality required significant effort. We had to ensure the prompts not only triggered the correct backend actions but also delivered a seamless and intuitive experience for our customers.

Additionally, ensuring the accuracy of data while respecting team boundaries required fine-tuning the JQL search. Balancing real-time performance with complex data processing was also crucial, leading us to optimize our algorithms to efficiently handle large data sets without compromising speed or accuracy.

Overall, this project was an enriching experience that brought both technical and problem-solving skills into play, resulting in a tool that effectively enhances Agile workflow.