M.S., M.A. · Founder, Logarchéon Inc. · Greater Tucson Area
Logarchéon builds group-invariant AI components — attention, representation, and arithmetic layers — for deployments where model leakage and uncontrolled inference are unacceptable. Built for AI model developers, regulated enterprises, law firms, and sovereign-AI programs that need evaluation under NDA.
The problem. Organizations deploying AI on sensitive data cannot yet show that a model’s internal representations resist extraction and misuse.
The edge. Graduate training in mathematics and prior training in theoretical physics, applied to group-invariant computation, where confidentiality follows from the architecture rather than from a wrapper.
The moment. Since 2 August 2025, the European Union’s AI Act (Regulation (EU) 2024/1689) has required providers of general-purpose AI models with systemic risk to ensure an adequate level of cybersecurity protection for the model and its physical infrastructure (Art. 55(1)(d)). [European Commission — AI Act]
Logarchéon Inc. (Delaware C-Corp, Tucson) is a one-founder research laboratory that licenses architecture IP and delivers evaluation builds of CEAS, GRAIL, and MIA. It is an IP-first operation: what it produces is patentable invention, working prototypes, and claims tested under a reproducible harness.
Operating model. Human founder sets direction and standards. Recursive AI agent stack executes derivations, simulations, tests, and technical writing. Public materials are non-enabling summaries; complete technical materials available under NDA and export-control compliance.
The core AI research program is grounded in the ASI Seed Architecture (U.S. provisional application 64/067,703, 2025): a seed architecture for artificial superintelligence combining global coordination (CEAS), causal reasoning (Ψ), and invariant geometric representation (GRAIL).
The four primary research systems—CEAS, GRAIL, MIA, and the Ψ-operator framework—are each summarized on dedicated pages; technical detail is available under NDA. Its confidentiality properties follow from group-invariant computation.
Global coordination channel with thermodynamic control of attention via explicit inverse temperature β. Physics-first training stability. U.S. provisional application 63/813,617.
Group-invariant neural computation on curved manifolds. Cryptomorphic twin models; automorphic attention. U.S. provisional applications 63/773,441 and 64/067,703.
Replaces scalar arithmetic with invariant-first group operations. Geometric error correction; designed for 28–65 nm processes and RISC-V and FPGA targets. U.S. provisional application 63/901,369.
Causal operator framework: intervention and counterfactual reasoning with explicit, checkable identification criteria; constrained inverse design.
I founded Logarchéon Inc. to create the institutional structure for independent, disciplined research at the intersection of AI, mathematics, and physics—work that does not fit inside a single academic department and that requires the judgment layer to remain with a human founder who sets standards and determines what work matters.
I studied theoretical physics at National Taiwan University and National Dong Hwa University, ranking first in my cohort at NDHU, before pursuing advanced mathematics at San Francisco State University (M.A.) and the University of Arizona (M.S.).
Licensing, joint development, and secure technical evaluations are available under NDA and export-control compliance. Peer review and third-party benchmarking of all AI architecture claims are explicitly welcomed. Conversations are welcome with law firms, regulated enterprises, technical reviewers, and sovereign-AI initiatives.
Evaluations begin with a 30-minute call and a mutual NDA; a technical packet follows within five business days.