Zero Hour

A system-debugging assessment platform for modern software hiring.

Comment

Inspiration

companies lose $300k+ per bad hire because leetcode doesn't tell you if someone can actually fix a system when it's crashing at 3am

students leave college knowing how to write code but nobody teaches what happens when that code breaks in production. like actually breaks.

i'm at teir 3. not iit, not nit. and i kept thinking why does nobody just teach the real job

so i built it

What it does

zero hour drops you into a broken production system and watches how you debug under pressure

live crashes. broken API. race conditions. the whole mess. you get a terminal, an editor, and a timer. we track how you think, not what you memorized.

it's not about pass/fail. it's about the behavioral data — how you isolate bugs, your debugging process, what you do when you're stuck. that's what companies actually need to know.

How we built it

backend: node.js/express on aws ec2 (mumbai region, t3.small), docker sandboxes for code execution so people can't break the actual server

ai: claude via aws bedrock generates the broken environments — backend services, database configs, deployment stuff, all realistically cracked

frontend: react on vercel, monaco editor for coding

real-time: websockets for terminal output and logs

db: postgresql for storing challenges and sessions

built alone. deployed to actual production. real infrastructure.

Challenges we ran into

docker isolation was brutal — keeping websockets stable while sandboxed code runs was harder than i thought. connections kept dropping, processes hanging. rebuilt the execution layer twice.

aws bedrock token limits — generating full production scenarios hits caps fast. had to chunk it and cache templates.

balance difficulty — early version was too hard. people just gave up. had to find the line between "realistic chaos" and "people actually finish". still figuring this out tbh.

what data matters — capturing the right behavioral signals without being invasive. what's useful for hiring vs what's just noise.

Accomplishments that we're proud of

actually deployed — not a hackathon demo. real system running on ec2 with proper docker sandboxing.

first completers — people finished challenges and said they learned more in 2 hours than weeks of leetcode. that hit different.

realized it's not ed-tech — companies will pay for hiring signal, not students paying to practice. changed the whole business model.

built solo as a 1st year — no co-founder, no tier-1 college brand, just shipped.

What we learned

technical: docker networking is hard. websocket reliability matters more in production. ai generation needs guard rails or it makes impossible scenarios.

product: too realistic = people quit. you need a difficulty ramp. also the process data is the actual product, not the pass/fail.

business: this isn't a learning tool. it's hr-tech. the market is companies paying to reduce bad hires. that changes pricing, positioning, everything.

personal: building alone means you ship fast but debug slow. every bug is yours. also aws bills add up when you're running sandboxes 24/7.

What's next for Zero Hour

more challenges — need 10-15 scenarios across different stacks (frontend, backend, devops, databases)

analytics dashboard — show companies the insights. how someone debugs, their process under pressure, where they got stuck.

b2b pivot — stop selling to students, start selling to eng teams as a hiring filter

workflow integration — make the cert travel with the developer (linkedin, github, resume)

targeting yc fr. this is a $300k problem for every tech company. if zero hour reduces bad hires that's a real business.

Built With

Share this project:

Updates