Skippy AI

Provide Jira users with different AI Providers to choose from to speed up repetitive activities, while being transparent about the data shared in the process and allowing them to stay in control.

Comment

Inspiration

Learn about the capabilities in offer when integrating with different AI Providers, and eagerness to see in practice both the challenges of this new world, as well as the dreams that could be achieved. Getting also to know how to build Atlassian Forge apps, and discovering the process and the challenges, was a big driver in starting this project.

What it does

The Skippy AI app is a Jira app, that allows the Jira users to speed up these 3 routine tasks: The creation of Story subtasks, the creation of Acceptance Criteria and finding possible fixes and solutions. The app does the above by integrating through our custom backend into 2 AI Providers for now: Google AI (PaLM2) and Open AI (ChatGPT-3.5). By using the capabilities of these two providers, the app is able to generate different propositions that the user can choose from. If the users are not happy with some of the generated subtasks or suggestions, they can easily request new suggestions through the app. The app is also clear about using AI Providers, and allowing the customer to choose whichever they prefer. The app also makes sure to allow the user to see what is the data that will be sent out to the AI Provider, and the user can choose to make changes to that before sending (like removing sensitive data for example).

How we built it

Process

After registering our interest in the Hackathon and got access to the supporting materials, we initially started by learning about building simple Forge apps through the tutorials provided, to understand bit more the challenges and possibilities there. Then we did few simple tutorials and POCs with different AI providers to see how we can integrate with these, and again understand how we can use them Once we had a bit of an idea of what can be achieved, we then tried to find an idea for the app to build through a brainstorming session. Finally, we decided on few features to build. We broke down the work to simple tasks, and tried to build it like an Agile project: 1) We first built the backend services that we used to communicate with the AI Providers. For that we used Spring Boot apps hosted on Google Cloud Platform as Serverless instances. 2) Then we built the UI, Frontend and Forge app features and integrate these with the backend services we created. For that we used UI Kit 2 and best practices as documented in the Forge wiki. We also tried to keep the user in mind and simplify what we could for now, while being transparent on the data used.

Challenges we ran into

One major issue was the latency from Open AI's ChatGPT-3.5, as some responses were taking way too long, even after extending some of the timeout configuration. While ChatGPT-4 could be faster, the pricing is bit too high for a startup to be able to provide that for Jira users. Thankfully being able to integrate with different AI Providers allowed us to offer the user other options that could be faster for now.

Accomplishments that we're proud of

1- First one is definitely getting over the line! It is always great to be able to achieve some of the goals we set up initially and be able to submit this project at the end. 2- Learning, learning and more learning: From getting hands on experience integrating with different AI Providers, to discovering a whole new world of building app through Atlassian Forge, plus all the unforeseeable issues, this has been a great journey where we learned a lot, for which one can be both proud and thankful to be part of it.

What we learned

In summary, we learned:

  • How to build Atlassian Forge apps and what possibilities are offered there, and especially what more we can build as new features are added.
  • How to integrate with different AI Providers, mainly Google AI and Open AI, how to communicate and get reasonable responses from an AI for different challenges.
  • What is Responsible AI and how to build apps that follow its principals.
  • In addition and tech stack wise: --> As mainly a Java backend developer, I learned a lot about using React and Node.js in this project :) --> Had the chance to also spend more time building backend services with Spring Boot --> Learned more about Google Cloud Platform services, mainly Cloud Run and API Gateway --> How to secure access to apps and services.

What's next for Skippy AI

This is hopefully just the start, we intend to:

  • Extend some of the services offered by the app in Jira, especially around proposing a solution, we would like to make that more useful by integrating with AI Image generation to get actual design models to the users.
  • Integrate with more AI Providers, the next we're targeting is AWS Bedrock, to give the users even more options.

Built With

Share this project:

Updates