Conjure for Confluence

Inspiration

Confluence is a great resource to keep important information, but for companies with hundreds of pages, knowing where to look for the answer to a single question can sometimes take ages. Enter Conjure, an AI system that knows everything you want it to know about your knowledge base.

Building a chatbot from scratch can be a daunting task, even for seasoned developers. Conjure allows you to build intelligent chatbots that are trained on your company's Confluence pages. It allows users to quickly and easily do what should typically take a few days in just a few minutes.

What it does

There are two components to Conjure:

1) Database Upsert Tool: This is a native Forge app that allows you to add any Confluence page to a Pinecone database. For large amounts of data, vector databases such as Pinecone are a must-have for any intelligent AI application. This tool will also scan the body of the page for any potentially sensitive information before it uploads the document. This Responsible AI practice allows companies to conform to ever-evolving privacy policies.

Pinecone is also now fully HIPAA compliant so using Conjure will allow you to conform to any strict regulations that may pertain to your industry.

2) Chatbot Interface: you can embed your very own chatbot into any webpage, either as a full page chatbot or a popup chat bubble. This chatbot is connected to your Pinecone database where it pulls information from the Confluence pages (which are now converted into vector embeddings to make it as fast as possible).

How we built it

Conjure is entirely written in JavaScript and powered by several frameworks and APIs.

It uses Forge to provide a quick data upsert tool for any Confluence document. There is a custom JS algorithm that will parse the text for any mentions of API keys, tokens, and secrets to ensure that the data about to added to the database doesn't contain sensitive information. The backend uses the Open AI embeddings API to generate vector embeddings which are then stored in Pinecone.

The chatbot interface was built using Flowise AI, which is based on Langchain JS. I also created a chatflow template specifically for Conjure, so you can get up and running in no time at all. All you have to do is connect your credentials for Open AI and Pinecone to have a fully functioning chatbot trained on your company's data.

Challenges we ran into

I entered last year's Codegeist (but did not win), so I was familiar with Forge but it had been a while since I used it, so I had to re-familiarize myself with the platform. I think the greatest challenge was figuring out a solid idea that I believe hasn't been done yet. There were also many instances of trial-and-error with figuring out where to place each component in the process. Conjure is a relatively complex app, so it took careful planning over the course of a few days to get it to run smoothly.

Accomplishments that we're proud of

This is an app that can be used to create intelligent chatbots who are experts on specific company data. Human Resources teams can use this to easily know which company policies need to be followed for specific situations. Engineering teams can use this to be quickly brought up to speed on complex problems that need solving. Finance teams can use this to gain a greater understanding of financial analytics. The possibilities are endless.

What we learned

I learned more about the Forge platform and reading through the blogs, documentation and discussions has very much inspired me to think more about the endless use cases for AI-powered Atlassian apps. The future of work is now, and we can use this technology to become so much more productive and efficient.

What's next for Conjure for Confluence

I think this app could be expanded to Jira in order to create a tool that could help developers troubleshoot issues that may be similar to bugs that have already been solved. This could be used to create an intelligent chatbot that could generate code that is specific to the tools and frameworks that a company is already using.