Venia Veselovsky and Theo Summer turned $35 into over $1.9 million on Kalshi, the event prediction market, using the AI forecasting system they built. That same system, a collection of independent AI agents they call Preseen, also placed first in the Metaculus Cup last year, beating over 1,000 human and machine forecasters [Waypoint]. Now, the Y Combinator-backed team is selling the engine behind those wins to institutions that need to see what's next.
Preseen's product is a calibrated probability. A client hands the system a question with a deadline, and within fifteen minutes, it returns a percentage chance that updates as new information arrives [South Park Commons]. The technical wedge, according to investor South Park Commons, is the use of multiple independent agents reasoning from primary sources, whose outputs are then aggregated to reduce correlated error [South Park Commons].
The Panel of AI Analysts
The core of the bet is an ensemble. Each AI agent within Preseen's system is prompted to reason independently, pulling from different data streams and analytical frameworks. Their outputs are then reconciled into a single, continuously updating probability. Beyond the Kalshi windfall, Preseen's system was the top-ranked forecasting bot on Metaculus and placed third in the Spring 2026 Metaculus Cup [Waypoint].
The early customer profile is the institutional decision-maker staring down high-stakes, long-tail risks. Jordan Berman, who has been building the product, frames it as building "the best AI forecasting agent for decision makers" and has publicly invited organizations to explore its work [LinkedIn].
Why South Park Commons Wrote a Check
The investment case rests on a technical differentiator in a field crowded with single-model point solutions. In a blog post explaining its investment, South Park Commons highlighted the multi-agent architecture as the key insight [South Park Commons]. The backing from Y Combinator in its Summer 2025 batch provided further validation and runway [LemStudio].
| Founder | Role | Key Background |
|---|---|---|
| Venia Veselovsky | Co-Founder | Built the top forecasting bot on Metaculus [Waypoint] |
| Theo Summer | Co-Founder | Co-created the Preseen forecasting system [Waypoint] |
| Jordan Berman | Builder | Leading development of the AI forecasting agent [LinkedIn] |
The Enterprise Pivot
Public leaderboards are one thing. Selling six-figure software contracts to risk-averse institutions is another. The company's next twelve months will be defined by its ability to convert its competition prowess into commercial traction. The risks are:
- Market education. The product is a probability, not a dashboard.
- Benchmarking. In a private enterprise setting, there is no public leaderboard to prove Preseen is beating its rivals.
- Category creation. There is no established software category for "AI forecasting agents" in the enterprise stack.