AI Competition Radar

A live radar of online, globally-open AI competitions, agent challenges and hackathons — upcoming, in progress, or ended within the last 7 days.

🔗 Live site: https://nobodyme.github.io/hackathons-online/

How it works

Layout and data are deliberately separated:

File	Role
index.html	Page shell / structure
styles.css	Design, responsive layout, light + dark themes
app.js	Fetches competitions.json and renders cards, filters and sort live in the browser — it recomputes status and countdowns from the current date, so the page stays accurate between refreshes
competitions.json	The only file that changes on a refresh — the competition list
assets/fonts/	Self-hosted Archivo (variable woff2) — no external font dependency
.github/workflows/deploy.yml	Publishes the site to GitHub Pages on every push to main

The visual style is Neo-brutalist: a butter-cream ground (#fff5db), chunky Archivo set heavy, 3px black outlines with hard offset shadows, zero corner radius, and bright color-coded status badges (orange accent, yellow = in progress, mint = upcoming). Design tokens live at the top of styles.css — restyling is a matter of editing those variables.

Because the page renders competitions.json client-side, refreshing the data is just a matter of overwriting that one file — the layout never has to be regenerated.

competitions.json shape

{
  "generated": "YYYY-MM-DD",
  "items": [
    {
      "id": "stable-kebab-slug",
      "name": "…",
      "platform": "Devpost",
      "url": "https://…",
      "type": "agent-challenge | coding-challenge | hackathon | benchmark",
      "format": "online | hybrid",
      "location": "Online",
      "start": "YYYY-MM-DD",
      "end": "YYYY-MM-DD",
      "reg": "YYYY-MM-DD",
      "prize": 10000,
      "prize_kind": "cash | mixed | non-cash | unknown",
      "prize_note": "e.g. split across tracks",
      "summary": "One sentence, ≤160 chars.",
      "conf": "confirmed | approximate | unknown",
      "eligibility": "Open globally, 18+",
      "audience": "all | students",
      "tags": ["agents", "llm"]
    }
  ]
}

start, end and reg are all required — no date field is ever null. A competition whose start or end date cannot be established is left out of the file entirely rather than published with a null: an entry you cannot plan around is not worth a row.

Many competitions have no registration step separate from submission (ARC Prize, MLH Global Hack Week, most Devpost hackathons). For those, reg carries the submission deadline, which is the same as end. So reg == end means "no separate registration cutoff — you have until the deadline"; reg < end means there is a real earlier cutoff to watch.

Rules the page enforces automatically:

• Retention: entries whose end is more than 7 days in the past are hidden.
• Status is derived from dates (upcoming / in-progress / completed).
• Audience: "students" marks competitions restricted to students/youth. These are hidden by default (with an "N student-only hidden" note); ticking the "I'm a student" checkbox includes them. Detection uses the explicit audience field when present, otherwise a scan of the eligibility text — so it works whether or not the data is tagged.
• Prize pools are shown as advertised — non-cash pools (credits/hardware) are flagged.

Automated refresh experiment

The content is curated by Claude on 3 day schedule, fully AI curated and pushes to this branch which then publishes the github page. Let's see how useful this experiment turns out to be.