Skip to main content

Command Palette

Search for a command to run...

🛡️ SafeUpgrade – My First AI Agent (Hackathon Project)

Published
2 min read
P

Versatile DevOps engineers. Well-versed with DevOps Tools and Cloud Services.

Why this project?

This project was built as part of the “AI Agents Assemble” hackathon.
The goal was not just to build automation, but to understand what an AI agent actually is and how it reasons.

How I selected this idea (real incident)

On Friday night, 5th December, around 9 PM, we received a Slack message from our CTO with a link to a Next.js security advisory:

A critical vulnerability was identified in the React Server Components (RSC) protocol.

Most of our production projects were using Next.js, and suddenly upgrading dependencies became risky but unavoidable.

That moment made me think:

“What if upgrades could be analysed and decided safely instead of blindly updating versions?”

That’s how SafeUpgrade was born.

Reference Link

What I built

I built a SafeUpgrade AI Agent that:

  • Scans dependency versions

  • Understands major vs minor upgrades

  • Blocks risky upgrades (major, canary, incompatible)

  • Applies safe minor upgrades automatically

  • Explains why something was upgraded or skipped

The project has two parts:

  • CLI Agent (Docker-based) for real DevOps workflows

  • Vercel UI Agent to visually demo agent reasoning

This is not just a script — it behaves like an agent that decides.

GitHub -

YouTube Demo -

What I learned (the most important part)

This is the part I’m most proud of.

  • I had zero AI knowledge

  • I had almost no JavaScript / Next.js knowledge

  • I had never built an agent before

  • I had never deployed anything on Vercel

  • I had never recorded a demo video or used Cline CLI

During this hackathon, I:

  • Understood what makes something an AI agent

  • Learned how agents reason, decide, and act

  • Built confidence working outside DevOps comfort zones

  • Learned Next.js App Router, API routes, and deployments

  • Used Cline CLI as the main agent interface

What I achieved

  • 🥇 My first hackathon project

  • 🤖 My first AI agent

  • 📦 My first agent-based DevOps tool

  • 🚀 My first Vercel deployment

  • 🎥 My first demo video

  • 🧠 A huge mindset shift about AI and agents

Final thoughts

This hackathon taught me one thing clearly:

You don’t need to “know AI” to build AI agents —
you need curiosity, real problems, and the courage to try.

This is just the beginning.


Thank you!🖤

Poonam Pawar — Github , LinkedIn , X

T

This SafeUpgrade AI agent is brilliant! Ὦ1️ The fact that it came from a real incident makes it so much more meaningful than theoretical hackathon projects.

I love how you"ve approached AI agents as reasoning systems rather than just automation. That"s the key insight - agents need to understand context and make decisions, not just execute scripts.

The upgrade safety problem you"re solving is massive. I"ve been in those Friday night situations where you"re torn between deploying fixes and risking weekend outages. Having an AI agent that can assess risk factors and suggest the safest approach could save so many teams from stress and downtime.

A question from my experience with voice-driven workflows: how do you handle the agent"s communication back to humans? Can it explain its reasoning in natural language when blocking risky upgrades?

Fantastic hackathon project with real-world impact! Ὠ0