Reasoning models that discover what humans miss.
We build machines that reason. The market is our litmus test: adversarial, nonstationary, and indifferent to a confident answer that turns out wrong.
Scale alone does not get there. Edge comes from architecture. We compose explicit memory, causal discovery, adversarial debate, and calibration into one reasoning cycle. The combination is the work; no single piece is.
Every position carries its reasoning, its calibration, and a decay clock.
Henri Francois leads SumLogLabs. He has researched machine reasoning since 2019. Jason Calacanis backed his first AI company when he was 21. Since then, he has built AI systems in finance.
We will build machines that think. The market will tell us when we have.
Our claim is that edge now comes from architecture, not scale. Three results in the recent literature converge: reasoning models collapse past a complexity threshold, chain-of-thought is brittle off its training distribution, and long-horizon memory is unsolved off the shelf. The response is compositional: eleven components, each anchored to the literature, assembled into one reasoning cycle.
Download the thesis (PDF) →A technical synthesis of the architecture, drawn from the research library. Correspondence: sumloglabs@gmail.com.
