Inspiration

Hackathons reward the team that picks the right thing to build - not the team that codes the longest. Most people still waste the first hours guessing themes, skimming Devpost, and recycling chatbot ideas. HackWin came from wanting a strategist, like a member of the team that does the research for you and returns one concrete product to ship, shaped by what judges actually care about

What it does

You enter a hackathon name and URL. HackWin researches sponsors, prize tracks, judging criteria, and past winners, then invents a single original product idea - not a just a project for a hackathon, a real thing to build. Each brief covers mission, target user, core loop, how it works, UI, screens, tech stack, demo plan, and a weekend schedule. Generate again and you get a new concept

How we built it

HackWin is a Next.js app with a research pipeline that searches and scrapes Devpost/official pages, year by year winner queries, and prize tracks into a structured dossier. An OpenAI model turns that into one detailed product brief. A perfection loop scores specificity and concept novelty, then retries with a forced new domain until the idea is concrete and different. Without an API key it falls back to demo mode

Challenges we ran into

All of the functions in the program kept failing in different ways. Early versions repeated the same concept under new names, so we had to ban prior missions/loops/users, not just titles. GitHub Pages looked like an easy deploy but can’t run the Next.js API, which forced a real host like Vercel. Keeping ideas specific enough to build from, without turning into a canned idea bank, was really hard to create

Accomplishments that we're proud of

A working end to end loop: name + URL --> research --> unique, buildable product brief. Ideas that change in concept, not just the name. Research that generalizes across events, plus a UI that puts the idea first where it belongs

What we learned

Winning strategy is taste matching: criteria and past winners are signals for what to build, not the product itself. Novelty needs concept fingerprints, not title lists. Scraping the open web is messy; graceful gaps and retries matter

What's next for HackWin

Richer research (official rules PDFs, Discord announcements), tighter demo scripts tied to the product’s before/after, team-size and stack constraints in the brief, optional export to a build checklist or Notion/Linear, and a hosted version with keys in the environment so anyone can get a live strategy in one clic, AI generated image of the product

Built With

Share this project:

Updates