Wizard

Main page
Main page with some alerts sent
Alerts received by a team
Using Atlassian AI with Alerts
Ask wizard, to directly enter meeting discussions and notify
Add labels to categorize alerts

Inspiration 😯

Inter-team communication shouldn't be inter-species communication, Let's keep it human! 😃

Picture this, you're working on a project while the entire requirements are still not finalized, only to find out that something you worked on for a week is no longer needed. Or, you're working in an Ops team where suddenly one day things behave strangely, only to find out that there was some change in a library you depended on. Or, you're a developer and an API call causes things to break due to a changed interface!

The main theme of all these problems is poor inter-team communication.

Our vision is to build a product that takes any inter-team communication and sends information to teams that need to hear it!

What it does 🕵️‍♀️

We take information from meeting transcripts (since this is the largest source of inter-team communication), extract important details identify all the stakeholders with Compass components, and send alerts to them.

In short, it takes important details from any meeting and sends them out to all developer teams that need to hear them!

How we built it 👷

We collected the required data from Atlassian Compass and from Confluence through Forge APIs. We then used a prompting technique called "Few-Shot prompting" to create a robust prompt that generates alerts out of the meeting transcript provided. This prompting not only extracts important information but also classifies the alerts to the Compass component affected. It also classifies general updates into the "General" Label (if added by the user).

Technology we used:

Forge custom UI for the user interface
Forge APIs to:
- Get compass component information and confluence page information.
- Update confluence pages with alerts
Open AI APIs to access the GPT3.5 model

Challenges we ran into 😮‍💨

Making a robust prompt for the few-shot technique was quite challenging, we also tried the OpenAI functions API but they weren't giving reliable results due to which we had to write a parser!
There were certain limitations with the Atlassian UI Kit, such as the behavior of setInterval and the UI Kit table not being customizable, due to which we needed to shift to the custom UI.
Making the presentation with all the different animal characters with no background in graphics was an interesting challenge. We have newly gained respect for people doing this! 🙇

Accomplishments that we're proud of 😎

Summarisation, generating meeting notes, Document Q&A, etc are common angles in AI. What we were really looking for are different angles at which we can use information from Atlassian products and tackle a real-world problem. We were surprised to find out how perfect Atlassian Compass was for this use case and we're proud of how we're using the Atlassian ecosystem.
Robust few-shot prompt design
Creating a product that we would use since it solves a problem that we and many others have personally faced.
Migrating to Custom UI from UI kit in 2 days! 💪
Animal character design in the presentation! 🐶

What we learned 👩‍🎓

Atlassian Forge Custom UI and APIs
New angles to look at how AI can be applied to applications
State-of-the-art prompting techniques, OpenAI functions API
Intricate use-cases of Javascript promises 😵

What's next for Wizard 🧙‍♂️

There's a lot in store! Currently, we are only using meeting information to bridge inter-team communication, we can expand this to take information from Jira and BitBucket to mitigate any inter-team communication gaps. This will also be a seamless integration with all Atlassian products. In the short term, we can also send alerts on compass announcements and make the Ask-Wizard feature more robust so that developers can ask wizards about the downstream impacts of any change that they are planning (Currently Ask-Wizard mainly works for dialogues)