DEV Community

Cover image for 🧠 How I Won a Hackathon With AI as My Entire Dev Team
Hemanth Babu
Hemanth Babu

Posted on

🧠 How I Won a Hackathon With AI as My Entire Dev Team

#webdev #vibecoding #hackathon #githubcopilot

🧠 How I Won a Hackathon With AI as My Entire Dev Team

TL;DR:

  • 🏁 Two apps, built in parallel, done in 45 mins
  • 💬 ChatGPT acted as my prompt engineer
  • 🤖 GitHub Copilot (Claude Sonnet 4) handled everything from UI to backend to test scripts
  • 🛠️ I just orchestrated: feeding prompts, reviewing code, nudging when needed
  • 🏆 Finished first — still waiting on the prize 👀

The Setup:

Internal hackathon. Two app-building challenges.

Everyone geared up to use GitHub Copilot.

I decided to do what I do best:

Let the AI agents take the wheel, while I just asked good questions..

📋 The Hackathon Brief

The event was part of an internal initiative on AI-assisted development — specifically around GitHub Copilot. After a few sessions on "Community of Practice," the hackathon was announced as a way to put those ideas to the test.

Two challenges. Everyone gets the same brief. Fastest (and cleanest) implementation wins.

🔧 Challenge 1: Shopping Cart

  • REST API for automobile parts (search, detail, pagination)
  • Shopping cart: add/remove items, calculate totals
  • Frontend: list view, product detail view, filters
  • (Bonus): Price slicer, modern UI

📋 Challenge 2: Interactive To-Do List App

  • Add, edit, delete tasks
  • Categorize by type (work, personal, etc.)
  • Due dates, reminders, priorities
  • Search and filter by any field

🧙 Enter the Prompt Engineer

I didn’t dive into code — I fired up ChatGPT.

I keep a dedicated chat I call my Prompt Engineer — it helps me generate perfect prompts for:

  • Building full apps
  • Debugging errors
  • Generating role/personality setups (e.g. for ChatGPT agents)
  • Image generation (like with Sora)
  • Styling or scripting tweaks — you name it

So naturally, I gave it both challenge descriptions and asked:

💬 “Give me a prompt for Copilot to build this functionality (Challenge 1). I like Python, but suggest the best stack.”

Then the same for Challenge 2.

In return, I got two perfectly structured prompts — ready to feed into GitHub Copilot’s agent chat.

My job? Just passing messages between two AIs.

📊 Here’s how the flow looked in practice:

AI Orchestration Flow

From requirements to results — one prompt at a time.

⚡ Copilot (Claude Sonnet 4) Gets to Work

I pasted the prompts into GitHub Copilot Chat (Agent Mode) — which runs on Claude Sonnet 4 — and sat back.

And just like that, my AI dev team clocked in. And then the magic began.

✅ The To-Do List App: Done in 10 Minutes

Copilot spun up:

  • Full frontend and backend
  • Batch scripts to launch it
  • Markdown documentation (usage, features, setup)
  • Even test scripts

I didn’t write a single line of code.

It even opened a browser window inside VS Code to preview the app. Don’t ask me how.

🛒 The Shopping Cart: Completed in 45 Minutes

While the To-Do app wrapped itself up, the Shopping Cart took a bit longer — but not by my hand.

Copilot generated everything:

  • API endpoints
  • Paginated product listing
  • Cart add/remove
  • Total price calculation
  • Markdown docs, test files, batch scripts — the full package

There was a small hiccup: clicking a product didn't show its detail view. I pointed it out, and it fixed itself.

Another bug threw an error. I didn’t even read it — I just asked ChatGPT Prompt Engineer to generate a fix prompt for Copilot. Boom. Fixed.

Honestly, it might have wrapped up even faster if it didn’t insist on writing the README, tests, and startup scripts too.
But I wasn’t complaining — it was like having a junior dev who cares too much about clean handoffs.

⏱️ Total Human Effort: 45 Minutes, 0 Code, 100% Prompts

From problem statement to working apps — Copilot did it all.

The only thing I did manually was manage the flow:

  • Asked ChatGPT for smart prompts
  • Fed those to Copilot
  • Gave feedback and nudges
  • Watched the apps build themselves

It felt like prompt-driven programming — not software development as we know it.

🏆 I Finished First (Still Waiting on the Prize)

By the end of the session, both challenges were:

  • Fully working
  • Documented
  • Tested
  • Launch-ready

I submitted, wrapped up, and sat back while others were still midway.

“What is the prize?”

They said goodies are coming.

Still waiting. 🫠

💡 What This Experience Proves

  • AI isn't just co-pilot — it's a team, if you prompt it well
  • You can orchestrate entire apps with structured instructions
  • Prompt engineering is a real skill — and a dev superpower
  • You don’t need to write code to ship quality software

This wasn’t cheating — it was effective orchestration.

🔗 Want to Peek Inside?

Here are the links to both repos — explore the code, docs, and flow if you're curious:

💬 Also, the exact prompts used for Copilot are included inside the repo VibeCoded-Projects.

🔮 Final Thoughts

I used to think hackathons were about caffeine, chaos, and typing fast.

Now?

It’s about:

  • Clear thinking
  • Smart delegation
  • And knowing which AI does what best

This wasn’t "cheating."

It was leveraging the best tools available — and letting humans focus on orchestration.

The future isn’t just developers typing faster — it’s developers thinking better.

🚀 What’s Next?

This is just one example. I’ve been using the same AI-assisted workflows for client projects too — like building & deploying websites in a single evening. (Wrote about that too.)

There’s more coming — and more experiments to try.

Until then… keep prompting. The future is listening.

Top comments (0)