Inspiration

This month in January 2026, someone turned $32,000 into $436,000 betting on Maduro's capture just hours before it was announced. A month earlier, another trader made a million dollars "predicting" 22 out of 23 Google search results. This was ousted as a Google employee insider. Insider trading on prediction markets is happening constantly, and no one is watching. I built this project to help change that.

What it does

Argus is an AI agent that will monitor political prediction markets 24/7. It will detect suspicious patterns like unusual win rates, disproportional bets, and suspicious timings and flag accounts with full evidence. The AI services the pattern, and then it presents it in a digestible way for humans to judge.

How we built it

It is a Next.js and React frontend, Convex for the database, and Claude via AWS Bedrock for the pattern detection and the in-house Polymarket API to grab the market data.

Challenges we ran into

Getting the AI to explain why an account is suspicious and not just that it's suspicious as well as reliably showcasing the way the information in a way anyone can understand.

Accomplishments that we're proud of

Built a working detection system that already flags real suspicious account data with detailed evidence.

What we learned

AI judgement often beats rigid hardcoded rules. I started off by hardcoding thresholds like "flag if win rate above this percentage," but letting Claude analyze patterns holistically was a lot more effective in catching subtle insider behavior.

What's next for Argus

Hopefully expand to other markets, and maybe not even just prediction markets. We also want to add real-time alerts and build a network analysis to help catch coordinated trading movement.

Built With

Share this project:

Updates