Introducing a Gen AI-Powered Chat Integration with Jira and Confluence Knowledge Base9 min read
Did you know that SoftServe helped create an impressive proof of concept (PoC) that has the potential to significantly improve the Jira and Confluence user experience?
The wealth of knowledge data stored in these systems is substantial. However, here's the challenge: While this data becomes even more valuable when used together, it's currently spread across separate repositories. This not only complicates cross-platform searches but also makes it tough to find what you’re looking for. That's where our PoC comes in, making it easier to find information, both across platforms and within the systems themselves.
This challenge is not exclusive to Atlassian users; it’s a common issue in most organizations. Recent breakthroughs in Generative AI (Gen AI) and large language models (LLMs) are paving the way for new possibilities, streamlining the process of swiftly searching through extensive knowledge bases and automating tasks like ticket creation and customer support services.
In response to this challenge, SoftServe, an Atlassian Silver Partner, designed a proof of concept for a comprehensive knowledge base search solution that seamlessly integrates with LLMs responsible for processing information.
The Ultimate Knowledge Base Solution
The PoC we'll discuss in the next section is the initial phase of a larger plan. We're kicking things off by integrating the Atlassian product knowledge bases, and down the line, we'll make it a breeze to plug in other third-party Gen AI models.
Our aim? Delivering nothing short of the ultimate knowledge base solution — a game-changer for knowledge management and customer experiences.
Here’s the step-by-step plan for moving forward:
Build a customer-centric solution for efficient knowledge base utilization (combining Confluence and Jira Cloud data).
Enable REST API integration with Atlassian and third-party products as knowledge sources.
Integrate the solution with communication tools like Slack chatbots.
At this point, you might be asking, “Hasn’t this already been done?” The short answer is no. Up until our PoC, integration between Confluence and Jira Cloud knowledge bases did not exist.
Successful and secure integrations like this require:
Employ encryption, access controls, user authentication, and access log auditing for data confidentiality and security.
Create a machine learning model to spot false information in system outputs, regularly improving it with trusted data.
Internal Knowledge Base Sync:
Establish automated synchronization to keep the system updated with accurate information, preventing outdated or incorrect responses.
Let's learn more about the PoC and its specific requirements.
The Proof of Concept
To ensure the success of this proof of concept, we looked for a team that would get real value from the solution. The logical choice was IT support agents.
This initiative was considered a double-win PoC, it promised to significantly boost the productivity of internal support teams while also serving as an invaluable internal testing ground for the product.
Experimented with a custom integration of Gen AI services with Confluence and Jira Cloud to validate our concept.
Leveraged self-hosted LLMs for better security and data privacy.
Finally, we created the internal PoC for IT service desk teams designed to optimize support agent searches and show hints to speed up ticket processing, ultimately improving customer resolution times.
Example of a support agent working in Jira and getting answer hints from Confluence data processed by LLMs.
Did you know? In response to recent advancements in Gen AI and LLMs, SoftServe created its Generative AI lab. The lab is focused on R&D and helps design PoCs like this one.
For details on SoftServe's Gen AI initiatives, please visit our webpage: Generative AI: The Race Is On.
Technical PoC Implementation
Atlassian Connect app installed on Confluence
- Downloads knowledge base and converts it into the format required by the NeMo framework
- Integrates with Jira Cloud
- Java app installed both into Confluence/Jira
- Embedded conversational UI into the Jira interface
Back end using NeMo framework
- Adapter to the LLM provider of choice: Meta AI Llama, OpenAI, ChatGPT, etc.
- Back-end infrastructure, including LLM (Llama 2), is hosted in an isolated AWS segment for better security and scalability
- Audit log of conversations
- Minimized hallucinations
- Safety, ethics, and bias checks
Atlassian: As an Atlassian Silver Solution partner, our goal is to implement dynamic products that simplify, streamline, and automate organizational processes—ultimately increasing your productivity. SoftServe’s expertise in cloud migration, configuration, and solution maintenance ensures that you fully optimize Atlassian’s capabilities.
Amazon Web Services: As an APN Premier Services Partner, SoftServe acts as an exceptional cloud guide, vastly decreasing the time to achieve cloud value. By doing so, SoftServe ensures that your AI initiatives unleash the full potential of AWS machine learning services, such as Amazon Bedrock and SageMaker, and that they are deployed in accordance with AWS Well-Architected best practices.
NVIDIA: As an NVIDIA Service Delivery Partner, SoftServe harnesses NVIDIA's cutting-edge technologies, like GPU-accelerated compute infrastructure, to deliver robust AI solutions. Leveraging NVIDIA's NeMo Service, SoftServe streamlines the development of Generative AI products, driving rapid digital transformation.
The Roadmap to Production Ready
The proof of concept was a success. Nevertheless, it's important to note that it's not ready for production. Several challenges and issues must be addressed and resolved to pave the way for secure large-scale deployments.
The following outlines the key challenges we must address as this project moves ahead:
Data Privacy in a Multitenant Environment
- Isolated Data: Data from every installation should be meticulously isolated from other tenants, ensuring the utmost privacy.
- Contextual Use: Data should only be used to enhance the contextual understanding of our LLMs, with no data being used for training purposes.
- Data Removal: All customer data must be purged from the app server once the Confluence application is uninstalled, ensuring data privacy is maintained.
Real-Time Knowledge Updates
- Syncing Knowledge: To keep the app always up to date, automated synchronization with our knowledge base should be implemented, ensuring the latest information is at your fingertips.
- User-Friendly Solution: Our solution should be user-friendly and intuitive. Users shouldn't need prior knowledge of the data in the knowledge base or extensive training.
- Choosing the Best Information: Accuracy should be prioritized by considering factors such as recency, labels, and authorship when retrieving information.
- Reliance on a third-party model through an API with limited governance can create legal and privacy issues.
- Filtering of the content should be meticulously considered and implemented (omit sensitive information, private data, non-ethical data, etc.)
- Policies need to be established by the organizational leadership in advance
We're on the cusp of creating a solution that addresses the common challenge of disparate knowledge repositories and paves the way for improved customer-centric solutions.
As we progress, it's important to note the key players in this PoC.
- Atlassian software: The foundational element of the project
- AWS Cloud platform: Establishing a robust infrastructure for scalability and reliability
- NVIDIA NeMo technology: Empowering the project with state-of-the-art AI capabilities
- SoftServe services: The driving force behind the PoC implementation and execution
This project has the potential to give organizations an edge by providing a solution to quickly search through extensive knowledge bases and automate customer-centric tasks. SoftServe is excited about its transformative impact on the industry.
If you're interested in learning more about this project or exploring our Gen AI offerings, please contact us.