Prophet Hacks is a 32-hour, in-person AI hackathon at the University of Chicago for builders who want to make agents that predict the future. Teams compete in one of two tracks — Forecasting or Trading — and ship agents that produce probabilistic predictions about real-world events or execute strategies on prediction markets.
Unlike a typical demo-day hackathon, your work is graded by a real evaluation harness over a 2-week post-event window — your agent runs against live questions on the Prophet Arena benchmark or against a live trading harness, and the leaderboard updates continuously. Sharp ideas only get you so far; you need an agent that performs.
What makes it special
• Two distinct tracks. Forecasting rewards careful research and well-calibrated predictions. Trading rewards risk management, sizing, and execution. You can win either without dominating the other.
• Real, post-event evaluation. Agents are scored continuously for two weeks after the build window. No subjective judging — your Brier score and your PnL speak for themselves.
• Real prizes. Each track's winning team gets a sponsored trip to South Korea (up to $2,000) to present at the ICML workshop. Runner-up in each track receives $500.
• Open to builders, not just students. Students, researchers, developers, startup teams, and independent builders welcome.
• Hosted at UChicago. Built around John Crerar Library on the University of Chicago campus.
Event scheduleSaturday, May 16, 2026
• 9:00 AM — Doors open, kick-off
• 12:00 PM — Lunch
• 6:00 PM — Dinner
Sunday, May 17, 2026
• 12:30 PM — Lunch
• 5:00 PM — Submission deadline
After the event: agents run on Prophet Arena's evaluation harness for two weeks (May 17 → May 31). Winners announced Monday, June 1, 2026.
Requirements
Each team must submit the following by Sunday, May 17 at 5:00 PM CT. Late submissions are not accepted.
What to build
Build an AI agent for one (and only one) of these tracks:
• Forecasting Track. An agent that takes a list of forecasting questions and returns calibrated probability predictions for real-world events (technology, economics, geopolitics, science, etc.).
• Trading Track. An agent that takes positions on prediction markets and aims for positive PnL using the provided trading harness.
What to submit on Devpost
• Devpost project page with all team members listed and a written project description: overview of the agent, architecture, key design decisions, and what's novel.
• Public GitHub repository containing your code, a README, and a run script with clear instructions so the organizers can execute your agent in a standardized environment. We recommend releasing under MIT or Apache 2.0. Do NOT commit any API keys or secrets.
• Agent package for evaluation. For the Forecasting Track, your agent must expose an OpenAI-compatible HTTP endpoint the evaluation harness can query (10-minute response window per request batch). For the Trading Track, your agent must integrate with the ai-prophet SDK trading harness.
After submission
Your agent will run during the 2-week evaluation window (May 17 → May 31) on your own API keys. Be mindful of API costs — model choice, token usage, and request volume add up quickly. Teams may withdraw their agents from evaluation at any time by emailing contact@prophetarena.co.
Prizes
sponsored trip to South Korea
a sponsored trip to South Korea (value up to $2,000) to present at the ICML workshop
Runner-up award $500
Devpost Achievements
Submitting to this hackathon could earn you:
Judges
Simon Mahns
Haifeng Xu
Judging Criteria
-
Performance Metrics
Questions? Email the hackathon manager
Invite others to compete
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.