From poker bots to Wall Street: a Czech AI startup's rise

Czech entrepreneur Martin Schmid, together with fellow scientists Matej Moravčík and Rudolf Kadlec, turned a student experiment with a poker algorithm into the startup EquiLibre, which today moves billions of dollars through the markets every day and whose valuation, after a recent funding round, has reached half a billion dollars.

The story began back in 2011 in a student dorm room: Schmid and his friends wanted to earn some extra money by building an AI capable of beating people at online poker. At the time, computers had long since surpassed humans at chess, but in poker, because of the element of uncertainty and bluffing, even the best algorithms still lost to professionals. Gradually the amateur experiments grew into serious research — the group went to Canada, where they continued developing their technology at a university. The work was eventually bought by DeepMind, the research lab owned by Google, which opened the team's door into the wider world of artificial intelligence.

Now Schmid and his partners are building EquiLibre — a company its founder is adamant about not calling just a “trading firm.” In his words, it's a full-fledged, cutting-edge AI research lab on the level of Anthropic, OpenAI or DeepMind. The only difference is in the application: instead of text-based language models, the team trains its own foundation models using reinforcement learning — and immediately releases the resulting “agent” onto the market, where it trades at high frequency fully autonomously.

Schmid's most unusual admission is that the team itself doesn't fully understand why their neural network makes the trading decisions it does. “As long as the line on the chart is going up, we'd rather focus on making it climb even steeper than on figuring out why it's happening,” the entrepreneur explains with a smile on the Money Maker podcast. The team's main criterion is the profitability of the algorithm's decisions, not the interpretability of its logic.

According to Schmid, it was precisely this early start in applying reinforcement learning to stock-market trading that gave EquiLibre a head start over competitors: the company considers itself a world leader in this narrow but extremely lucrative niche — automated AI trading on financial markets.