Projects
A Bayesian runtime for probabilistic reasoning.
True intelligence is not recall; it is prediction. xRtm provides the simulation loop, snapshot vault, and rigorous scoring rules needed to train world models that anticipate the future.
Read the manifesto at xrtm.org.
Verifiable reasoning traces for high-stakes medical coding.
EvidenceCoder builds citation trails that tie each coding decision to source evidence, enforcing structured rationale paths that are auditable and resilient on long-tail cases where models often drift or hallucinate.
Base paper available here.
A self-correcting QA framework for conflicting knowledge.
Standard RAG retrieves documents but cannot resolve conflicts. OpenVerify is a QA system that identifies contradictions in retrieved contexts and generates "checking questions" to resolve the truth before answering.
Coming soon: Interactive demo.