AutoDev Guardian

An autonomous AI engineer that builds, secures, deploys, and continuously improves software—turning ideas into production-ready code with zero human intervention.

Comment

🚀 Inspiration

Modern software development is still heavily manual. Developers spend hours writing code, fixing vulnerabilities, updating dependencies, debugging pipelines, and optimizing performance.

We asked a simple question:

What if an AI system could not only build software—but also secure, deploy, and continuously improve it on its own?

Inspired by recent advancements in autonomous agents and the concept of "auto-research" (where AI improves its own code through iterative experimentation), we set out to build a self-improving AI software engineer.

⚙️ What it does

AutoDev Guardian AI is an autonomous multi-agent system that transforms a simple feature request into production-ready, optimized software.

Given a GitLab issue, the system:

  • ✨ Generates production-ready code
  • 🔧 Fixes outdated or vulnerable dependencies
  • 🛡️ Detects and patches security vulnerabilities
  • ✅ Runs tests and prepares a merge request
  • 🚀 Deploys the application
  • 🧠 Continuously improves the code using an automated research loop

💡 The standout feature:

A self-improving Auto-Research Agent that:

  • Modifies code iteratively
  • Runs controlled experiments
  • Evaluates performance
  • Retains only the best improvements

This turns the system into a continuous learning and optimization engine.

🏗️ How we built it

We designed a multi-agent architecture, where each agent has a specialized role:

  • Feature Builder Agent → Generates code from feature requests
  • Security Triage Agent → Detects and fixes vulnerabilities
  • Dependency Healer Agent → Updates and stabilizes dependencies
  • Auto-Research Agent → Improves code through iterative experiments
  • Deployment Agent → Handles CI/CD and deployment

🧠 AI Layer

  • Powered by Mistral 7B (Instruct) — an open-source LLM
  • Prompt-engineered agents with role-specific reasoning
  • Retrieval-Augmented Generation (RAG) for repo context

⚙️ Backend

  • FastAPI for orchestration
  • Agent controller managing workflows

💻 Frontend

  • Command-center style dashboard
  • Real-time agent status and logs

🔬 Research Engine

  • Docker-based sandbox execution
  • Iterative experiment loop (fixed-time runs)
  • Performance tracking and comparison

🔗 Integrations

  • GitLab API (Issues, Merge Requests, Pipelines)
  • CI/CD simulation
  • Security scanning tools (SAST / Trivy)

⚔️ Challenges we ran into

  • Agent coordination: Designing a reliable multi-agent workflow without conflicts
  • Model limitations: Optimizing Mistral 7B outputs with prompt engineering and RAG
  • Safe code execution: Building a sandboxed environment for experimentation
  • Measurable improvements: Defining meaningful metrics for the Auto-Research loop
  • Balancing realism vs prototype: Simulating real GitLab workflows while keeping it hackathon-ready

🏆 Accomplishments that we're proud of

  • ✅ Built a fully autonomous, end-to-end SDLC system
  • 🧠 Implemented a self-improving AI research loop
  • ⚙️ Designed a scalable multi-agent architecture
  • 🛡️ Integrated security + dependency + DevOps automation
  • 💡 Used open-source AI (Mistral 7B) for a deployable, cost-efficient system
  • 🎯 Created a compelling, real-world applicable solution

📚 What we learned

  • Multi-agent systems are far more powerful than single LLM pipelines
  • Clear role separation dramatically improves AI performance
  • Iterative experimentation (Auto-Research) is key to building self-improving systems
  • Open-source models can deliver strong results with the right architecture
  • The future of software engineering is autonomous + continuously improving systems

🔮 What's next for AutoDev Guardian

  • 🔄 Full real GitLab integration (live pipelines and deployments)
  • 🧠 Smarter Auto-Research with advanced optimization strategies
  • 📈 Reinforcement learning for long-term improvements
  • 🌐 Support for large-scale enterprise codebases
  • 🤝 Multi-repo and team collaboration features

🏁 Final Note

AutoDev Guardian is not just a tool.

It is a step toward:

Self-improving, autonomous software systems that can build, secure, and evolve on their own.

Built With

  • chromadb
  • docker
  • faiss
  • fastapi
  • github-actions
  • gitlab-rest-api
  • gitlab-sast
  • langchain
  • mistral-7b-(instruct)
  • ollama
  • react.js
  • trivy
  • uvicorn
Share this project:

Updates