The problem Kowalski Market solves

Kowalski Market is an on-chain prediction market where the live trading price of every event is its real-time probability. Three distinct features set it apart: (1) permissionless market creation — anyone can deploy a market with custom rules by posting a bond, with a curator validating before it goes live; (2) Kowalski, a voice- and text-controlled AI layer that lets users trade, place limit orders, navigate, and discuss markets in natural language while every transaction is signed by the user's own wallet; and (3) Central-Europe-first coverage of events legacy bookmakers and global prediction markets ignore — CE politics, weather, energy prices, regional sport, and culture.

Together these features unlock things today's prediction markets can't:

1. Niche markets get real liquidity. A user asks "Will US tariffs on European steel exceed 15% by Q3 2026?" - user get a prediction on the probability of an event of his/hers interest

2. Standard prediction markets and their interfaces aren't intuitive — most users don't understand order books or CLOB mechanics. Through Kowalski's agentic interface, beginners interact by simply talking, and the agent explains whatever they need along the way.

3. Research happens in the same flow. "Why did this move 8% in the last hour?" — the agent surfaces the news behind the move. Voice works hands-free.

4. CE coverage that didn't exist finally does — politics, weather, energy, regional sport, and culture, in the region's own language and context.

Challenges we ran into

The interesting part was designing the prediction market around the end user — improving on what existing prediction markets get wrong and rethinking how the trading surface should feel for someone who isn't already a crypto-native trader.

On the technical side, ENS resolution was returning stale or near-match records, so Kowalski would confidently answer about a different market than the user named — solved by exact-match first, fuzzy only as a confirmation prompt, plus a sanity check that the resolved market's question matches user intent before any action. The other big opportunity was making a CLOB and an AMM coexist so liquid markets get tight order-book spreads while freshly created ones aren't dead on arrival; we routed every quote through a single pricing layer that aggregates both sources, with the AMM as implicit counterparty when the book is thin — fillable prices on day one, tighter book-driven prices once real makers arrive.