AutoPipe

🚀 Inspiration

Creating CI/CD pipelines in GitLab is powerful but often daunting, especially for those unfamiliar with YAML syntax or job dependencies. We wanted to build a tool that lets developers focus on what they want to accomplish—not how to write the pipeline.

So we built AutoPipe, an AI-powered assistant that helps you build a GitLab pipeline from scratch, starting with just a natural language user story.

🛠️ What it does

AutoPipe guides users through the full pipeline creation process:

🧠 Understand a user story and generate initial GitLab Pipeline draft
🤖 Fill in missing details interactively with clarifying LLM prompts
📋 Summarize the complete pipeline in plain English
🖼️ Visualize pipeline structure (WIP)
📦 Show print out a production-ready .gitlab-ci.yml file

Users can interact via CLI or a web UI built with Streamlit.

🔧 How we built it

We originally planned for a CLI-only tool, focusing on Python + Click for structured commands. As the hackathon progressed, we realized that deployment and interactivity were required. With only a few hours left before the deadline, we pivoted and discovered Streamlit, which allowed us to rapidly build a user-friendly web interface.

The stack:

LLM Backend: Google Gemini Api
Core Logic: Python with Click and Pydantic
Web UI: Streamlit (built in a few hours!)
Deployment: Dockerized and deployed to Google Cloud Run
State Management: JSON-based pipeline model
Prompt Design: Templates that allow Gemini to ask clarifying questions and return structured YAML

🧱 Challenges we ran into

Adapting a CLI-only tool into a deployed web service under tight time pressure
Managing session state in Streamlit while ensuring LLM chat history consistency
Writing prompt templates that could generalize across many pipeline types
Build and Deploy this app to Google Cloud while we just have 2 hours left

🏆 Accomplishments that we're proud of

A working product that turns plain English into a GitLab pipeline
Dual-interface: use it in terminal or via browser!
Successful bring my first project online by deploying it to Google Cloud Run
Complete user flow: initialize → generate → refine → show
Built the entire UI and deployed it in the last few hours of the hackathon!