We're a small research team building AI systems that forecast real-world events, the kind of questions that shape policy, markets, and lives before anyone has good answers. Will AI replace software engineers by 2030? Will the US tip into recession? When does the war in Ukraine end? How many climate migrants will cross borders in the next decade?
Right now, human superforecasters still beat AI models on every major benchmark. But the gap is closing fast, and AI forecasters are projected to surpass them by 2027. We want to get there sooner.
Our work spans the full stack of what it takes to build a forecasting system that actually works. We fine-tune LLMs, build agentic systems, and push how models reason and retrieve information. We train for calibration, not just accuracy. We build reinforcement learning environments where models learn from resolved predictions. We backtest against years of historical data. And at the core of it all is deep research, because a forecast is only as good as the evidence behind it.
It's demanding work. It asks us to sit with ambiguity, to question our own assumptions, and to hold ourselves to a standard where being approximately right matters more than sounding precisely wrong. But it's deeply rewarding, because the world makes better decisions when it knows what's next.
AI forecasting is a new field. We think it will be one of the most important. If you want to work on it with us, reach out.