I built LocksBet to make it easier to see where the same event is trading across prediction markets, and compare the price you would actually get after accounting for fees and liquidity.

The site matches related markets on Polymarket/Kalshi, shows (after-fee) prices side by side, and lets users compare the underlying rules.

Currently, the site uses embeddings and hard gates on stuff like numerical range/date differences to find match candidates. An LLM then sanity checks the embeddings-based match candidates, and sends those that seem like true matches to human review. I then manually review match candidates and use the manually labelled data to improve our automated matching over time.

I'm working on adding more prediction markets as new platforms launch, with the goal of making it easier to find the best available price across venues. I'm also working on showing more complex relationships between events (e.g. one-to-many outcome matches).

It's been fun to work on and I'd love to get any feedback on the site and any other thoughts/questions.