MergeMind AI

Inspiration

Modern software development is slowed down by repetitive DevOps tasks like writing boilerplate code, reviewing pull requests, managing CI/CD pipelines, and fixing deployment issues. We were inspired to build a system that could act like an autonomous engineer — handling these tasks intelligently without constant human intervention.

The idea was to transform GitLab from a passive tool into an active, intelligent system that can think, act, and improve the development workflow on its own.

What it does

Our project is an Autonomous DevSecOps AI System that integrates with GitLab and automates the entire software development lifecycle.

It listens to GitLab events such as issues, commits, and merge requests, and then triggers a network of AI agents to:

• Analyze and plan tasks
• Generate code
• Review and improve code quality
• Perform security checks
• Run CI/CD pipelines
• Deploy applications
• Detect and fix failures automatically

In simple terms, it acts like a self-operating engineering team inside GitLab.

How we built it

We built the system using a modular, event-driven architecture:

• FastAPI as the central orchestrator
• GitLab Webhooks to receive real-time events

• Multiple AI agents for different tasks:

  • Planner Agent
  • Code Generator Agent
  • Review Agent
  • Security Agent
  • CI/CD Agent
  • Deployment Agent
  • Incident Response Agent

• LLMs (Claude/GPT/Ollama) to power intelligent decision-making

Each agent is designed to handle a specific part of the pipeline, and the orchestrator coordinates them to complete tasks end-to-end.

Challenges we ran into

• Designing coordination between multiple AI agents without conflicts
• Handling real-time webhook events reliably
• Ensuring generated code is valid and secure
• Managing long-running processes outside GitLab CI
• Balancing automation with control to avoid unintended actions

Accomplishments that we're proud of

• Built a fully autonomous DevSecOps pipeline
• Successfully integrated AI agents with GitLab workflows
• Created a system that can go from issue → deployment automatically
• Designed a scalable multi-agent architecture
• Demonstrated real-world applicability beyond just a demo

What we learned

• How to design and orchestrate multi-agent AI systems
• Deep integration with GitLab APIs and webhooks
• Challenges of automating real-world development workflows
• Importance of clear system architecture in AI-driven applications
• How to combine cloud and local LLMs effectively

What's next

• Add learning capabilities so the system improves over time
• Enhance security intelligence and vulnerability detection
• Build a visual dashboard for monitoring agent activity
• Support multiple repositories and team collaboration
• Optimize performance and reduce latency

Built With

• ai/llm-models-(claude
• and
• basic
• ci/cd
• docker-(optional)
• fastapi
• gitlab-api
• gitlab-webhooks
• gpt
• ollama)
• python
• render/railway-(for-deployment)
• rest-apis
• uvicorn