Building EduHub Solo: What Three Days on a New Stack Actually Taught Me

I'd been building EduHub for some weeks before H0 showed up on my radar. It started as a fairly ordinary problem: teachers spend hours either writing worksheets from scratch or hunting for ones that actually match their syllabus, and most AI tools that promise to help still need someone to manually check every question for accuracy. I had a working app on Supabase, a decent chunk of the product already built, and no plans to touch the database layer again.

Then I did the smart thing, and read the hackathon's rules. The primary backend had to be an AWS database, Aurora PostgreSQL, Aurora DSQL, or DynamoDB, and Supabase didn't count no matter how well it worked. So three days before the deadline I had to either migrate the entire thing, or drop out entirely.

This was also the first time I'd built something genuinely solo under real time pressure. Every call was mine: whether to migrate auth too or keep it separate, whether to risk DynamoDB's speed or stick with what my schema already wanted to be. I chose Aurora deliberately. My data is relational by nature, curricula link to skills, courses link to worksheets and questions, all through foreign keys. Denormalizing all of that into DynamoDB under a deadline felt like solving a problem I didn't have, just to look more "AWS native." Aurora mapped onto my existing schema with zero redesign, and that just sounded simpler for everyone (especially me).

The stack itself wasn't the hard part, eventually. The hard part was the order of operations: provisioning Aurora through Vercel's Marketplace integration, getting IAM auth working instead of falling back to static keys, re-pointing reads before writes so I could catch mistakes early. None of it was complicated in isolation. Doing all of it solo, while also making sure the actual product still worked end to end, is where the real learning happened.

What I didn't expect was how much I'd end up caring about this past the hackathon deadline. Somewhere in the middle of rebuilding the data layer, I kept coming back to the actual people this is for, teachers who don't have six extra hours a week to spend hunting for resources, and who currently have to choose between speed and trusting that the AI-generated worksheet in front of them is actually right. That's not a hackathon problem. That's a real one, and it didn't take long for it to stop feeling like a project I needed to finish in time and start feeling like something I actually want to keep building.

So that's where I am now. One hour off from submitting the product, and I'm actually a little excited to hop back onto VS Code in the morning (never thought I'd say this, ever.) Happy days!