Z-GPT - AI Assistant for Jira & Confluence

Z-GPT integrates AI into Atlassian ecosystem, enabling intelligent automation, document generation, and workflow optimization for Jira and Confluence users.

Comment

Inspiration

In the modern agile workflow, teams using Jira and Confluence often face three silent productivity killers:

  1. Writer's Block: Spending too much time drafting clear ticket descriptions, acceptance criteria, or documentation.
  2. Language Barriers: Global teams struggling to communicate technical details across languages.
  3. Missing Visuals: Documentation that lacks visual context because creating mockups or diagrams takes too long.

We built Z-GPT to be the ultimate "Sidecar Assistant" for these workflows. Instead of switching between five different browser tabs for translation, drafting, and image generation, Z-GPT provides a unified, secure, and local AI environment to handle these tasks instantly.

What it does

Z-GPT is a full-stack AI workspace that combines three powerful capabilities:

  • Smart Chat (Drafting Assistant): Uses local LLMs (TinyLlama/Mistral) to draft Jira tickets, summarize technical issues, and write Confluence pages.
  • Real-time Translation: Instantly translates text between languages, ensuring that comments on Jira tickets are understood by everyone, regardless of their native language.
  • AI Image Generation: Creates visual assets, mockups, and diagrams on the fly using Stable Diffusion, perfect for enriching Confluence pages without waiting for a designer.

How we built it

We focused on a privacy-first, local-first architecture to ensure sensitive project data never leaves the user's infrastructure.

  • Backend: Built with FastAPI and Python. We used SQLModel for a robust database layer (SQLite/Postgres) to manage persistent chat sessions.
  • Frontend: A responsive React SPA (Single Page Application) that provides a seamless, chat-like interface.
  • AI Engine:
    • Hugging Face Transformers: For running optimized LLMs locally on the CPU/GPU.
    • Diffusers: For the image generation pipeline.
    • Argos Translate: For offline, neural machine translation.
  • DevOps: The entire stack is containerized with Docker, making it easy to deploy alongside existing internal tools.

Challenges we faced

  • Resource Constraints: Running multiple AI models (LLM + Stable Diffusion) simultaneously on standard hardware was difficult. We solved this by implementing lazy loading and allowing users to toggle features via environment variables (IMAGE_ENABLED).
  • Streaming Responses: Providing a "ChatGPT-like" experience required implementing Server-Sent Events (SSE) in FastAPI and handling real-time stream parsing in React, which was complex to get right with error handling.
  • Cross-Platform Compatibility: Ensuring the app runs smoothly on Windows (PowerShell) and Linux/Mac required careful dependency management and path handling.

Accomplishments that we're proud of

  • Successfully integrating three distinct AI modalities (Text, Image, Translation) into a single, cohesive application.
  • Achieving a clean, production-ready codebase with comprehensive CI/CD workflows, automated testing (Pytest/Jest), and security headers.
  • Building a system that runs entirely offline/locally, offering a zero-cost alternative to expensive cloud AI APIs.

What's next for Z-GPT

  • Direct Atlassian Integration: Building a dedicated Atlassian Forge app to embed Z-GPT directly inside Jira tickets.
  • RAG (Retrieval Augmented Generation): Connecting Z-GPT to a Confluence knowledge base so it can answer questions based on existing company documentation.
  • Voice Interface: Adding speech-to-text for dictating ticket updates on the go.

Built With

Share this project:

Updates