The Story of AlBert and Codegeist Unleashed 2023

[#1]

Here we are again, Atlassian Codegeist, for the 2nd time. Forge and AI sounded just too familiar after we made it to 6th place in Codegeist 2020, and we could not resist having another go after 3 long years.

Ideation - How AlBert was conceived

I must say we actually did a very selfish thing - since there was no comparable app on the Marketplace we wanted to create AlBert for ourselves for a long time to speed up our creative content creation flow in Confluence and make it a helluvalot easier. So Codegeist was the perfect excuse to take a break from the daily grind and get C R E A T I V E.

AlBert’s birthing

Since we have played with Forge before, the technology was somewhat familiar, but BOY it has come a long way since mid-2020.

Forge does a very good job of integrating systems - and THANK GOD (or better even thank you, fellow developers giving Atlassian feedback) there is a Custom-UI in Forge.

Anyways - armed with some experience, the right tools, an API token for api.openai.com and the willingness to overcome any obstacle in our way (see below) - we went to slay the dragon.

Thanks to the fantastic work of Dugald the First of Atlassian , we were able to get started with AlBerts implementation straight away and saw some promising results within hours! AlBert's first cry!

 

AlBert the toddler

Now, with little AlBert being such a good little baby, it was time to take on teaching him to crawl, sit, walk - you know the drill. We created a simple pattern consisting of Forge custom-ui components, forge resolvers, a prompt engineering module and API module that allowed us to pretty much to teach AlBert to do anything we wanted!

But not everything goes that smoothly. AlBert kept tripping over that Forge-imposed 25s timeout threshold. OpenAI’s popular LLMs are definitely not the fastest in giving away their responses. So whenever AlBert had something to say, Forge’s ears were deaf.

As responsible parents, we launched a search for the fastest AND most comprehensive LLM out there. Turns out gpt-3.5-turbo-instruct is the Porsche of available models. Fast and no compromises. After hooking up AlBert with his first Porsche everything went super smooth.

BTW - Atlassian offered some help by increasing the Forge timeout threshold to 55s - but we wanted AlBert to be F A S T, otherwise - what’s the point, right? Once you drive a Porsche….

AlBert’s first visit to the playground

AlBert’s now really stable on his feet and learns new things every day - time to release him to a safe playground with a fence where only friends and friendly people hang out with their kids. So we are taking AlBert to Codegeist 2023, planning to sit back and watch AlBert play, and see how he does?

We’re confident we’ve done all we can as AlBert’s parents to get him ready for playday. Now it’s up to you, dear fellow Atlassian ecosystem dwellers, the judges and a lot of good luck to see our second child AlBert on the podium.

We have plans for AlBert

Codegeist is the perfect way to test out new product ideas against a huge audience. If you’re a product manager, Codegeist is finger-licking-awesome to get feedback early.

We’re pretty sure AlBert will do well, and we have big plans for how we want to further his education.

Our goal is to teach AlBert to connect easily to any domain, knowledge base, or documentation resource, including any Confluence instances or spaces. AlBert will be learning to use LangChain to tap into domain-specific knowledge in order to make content creation even more valuable and tailored to people that want to hang out with AlBert.

If you want to find out all about AlBert check out the intro video and our Codegeist Submission.

Ready to play with AlBert? Log into your Confluence cloud instance and open the AlBert share link in a new browser tab, install and let AlBert inspire your content creation process.