Frogs guarding the Well Maria from Drought titan: Defeating the "Heat and Water" Crisis with Proactive AI

Racing against the hackathon deadline, we pivoted our heavy PyTorch GraphSAGE model to AWS EC2 to build a proactive digital twin for drought prediction. Read how our team successfully engineered this complex backend architecture without resorting to any Loki-style mischief or chaotic shortcuts.

BPP2

Published Oct 11, 2026

Last modified Oct 11, 2026

Let me set the scene for you... 4 B.Tech 2nd-year CSE students. 3.5 days on the clock. A massive drought intelligence engine to build. And a cloud computing platform that we had absolutely zero business messing with.
Welcome to Team All Starz's journey of building CASCADE. Or as I like to call it: Jumanji: Welcome to the Amazon Web Services Jungle.
As the team lead, I took charge of the backend files right out of the gate. The plan was simple: build a PyTorch GraphSAGE model, host it on AWS, connect the Next.js frontend, and win. Easy, right? Wrong.
Before this hackathon, I was completely foreign to AWS. I knew it existed, sure, but I’d never built anything large-scale enough to actually need it. So, while Alisha and I were trying to figure out which AWS tools to actually use, I went on a frantic learning montage. I asked Google AI, interrogated ChatGPT, and basically made Gemini my personal cloud tutor. After hours of absorbing this sacred knowledge, I marched back to the AWS console like Chanakya returning for his revenge. I was ready. I was enlightened.
And then... AWS refused to let me sign in.
Every time I tried, it threw an error. It felt exactly like that trope in Final Destination—you think you’ve finally escaped the danger, you take a breath of relief, and then BAM, a massive error block strikes you with the reality that you were doomed from the start. Haar maanke, I hit up the AWS help center. They replied in two minutes: “We are facing high traffic and are processing requests in order.”
Cool. Awesome. The deadline is now racing past us like Max Verstappen on a straightaway, and I’m in a digital waiting room.
Enter the team effort. Alisha had worked with AWS once before, so she took the wheel. But alas, the AWS bouncer wouldn’t let her in either. We were quite literally locked out of our own backend. Here comes the absolute savior of our project: Alisha’s sister. She had a pre-claimed AWS account. We borrowed it, and finally... finally... our backend zip was posted on EC2. We fought a ridiculous amount of errors in the process, but the heavy lifting was done.
Now, we have 1.5 days left.
The division of labor at this point was... interesting. Anju and Kashish were deep in the trenches designing a beautiful UI/UX for the dashboard. 
﻿
Alisha was aggressively fighting AWS backend bugs. And me? The fearless leader? I was temporarily jobless. The backend was up, the frontend wasn't ready to connect yet, so I was literally sitting idle, casually chatting with Gemini about other hackathons and competitions. (Hey, a leader has to plan for the future, right?...)
Fast forward to the evening before the deadline. The frontend is finally done. It’s my time to shine again. I take the AWS endpoints, strip out our placeholder data, and wire the Next.js app to the live backend. It works locally. I deploy it to Vercel. I check the clock. It is 5:00 AM. I have been surviving on 3 hours of sleep for two consecutive days, operating purely on caffeine and vibes like I’m Chitti the Robot. I close my laptop and collapse into bed.
But the chaos wasn't over.
While I was dead to the world, Alisha was still awake. She was designing our landing page and squashing the final critical UI bugs. That’s when she realized my sleep-deprived brain had deployed the Vercel link using a forked repo... from Anju’s GitHub... which was set to private. Meaning my Vercel plan wouldn't let collaborators make any changes.
When Alisha’s crew faced this final, boss-level error, she didn't panic. As a true full-stack developer, she had her Tony Stark moment. She stepped up, created a brand new public repo, migrated the code, integrated the landing page, and with a single terminal snap... turned all those deployment errors to dust. She even volunteered to fill out the agonizing hackathon submission form herself. Absolute legend.
﻿
To top it all off, I handed our demo script over to Kashish. She improvised it, smoothed it out, and recorded a flawless voiceover that made our frantic 3.5-day sprint look like a polished, enterprise-grade startup.
Looking back at the Google Meets that stretched until 5 AM, the sheer pressure we felt as individuals, and the collective panic of the team... it was exhausting. But we survived the jungle. We built a proactive, AI-driven drought simulation engine in less than a week. For four 2nd-year students, CASCADE set a standard I didn't even know we were capable of.
Team All Starz, signing off. (And don't you dare ask me whether my AWS signing-in error has been resolved yet or not) 😕🥀