Portfolio / Connected pursuits
Our work connects the possibilities of computation, the control of intelligent systems, and the demands of consequential execution. These pursuits inform one another; they are not sequential stages.
What can a computation become, and which possibilities must remain open?
Formal foundations and engineering systems that connect computational intent to physical realization, without erasing choices the future may still need.
How should intelligence reason, learn, and acquire the authority to act?
Architectures that connect cognition, evidence, and learning while keeping consequential action subject to a distinct, accountable control plane.
How does frontier capability become work that meets real-world requirements?
Platforms that bring domain expertise, evidence, decisions, and accountability into the operating core of consequential work.
Realization defines the choices. Agency governs action. Execution supplies the requirements that sharpen both.
Research in the OpenComputation and realization / CRG
A formal theory of how computation becomes physical, developed into a system for preserving valid chip-design choices and governing safe commitment.
The theoretical contribution
CRG maps valid physical realizations of a computation and the joint constraints across movement, state, work, latency, replication, ownership, and hierarchy.
An option may be discarded only when no decision still ahead could tell that it was gone. The law makes premature loss of a viable design path testable.
The operational system registers alternatives, computes their obligations, and issues auditable, independently verifiable certificates for commitment and change.
Foundational research / Agentic intelligence
Evidence changes what an agent believes—and what it is permitted to do.
An AI agent can choose which checks to run, which evidence to seek, and when to stop reasoning. When those choices also shape its permission to act, cognition and governance become one coupled control problem.
The Agentic Brain connects fast inference, deliberation, verification, and protected authority through an agentic control plane. Formal results and executable finite-model validation examine the incentives this coupling creates, the evidence required for valid admission, and the boundary between permission and external effect.
· Preprint v1.0.0 · Open access
Value-of-computation analysis separates improved beliefs, changes in admissible actions, and their interaction.
Assurance must remain valid for the checks an agent can select. Complete logging alone cannot repair a verifier blind spot.
Single-use authorizations bind permission to the operation that actually runs. Faster competence does not automatically confer authority.
Product platform
Fab-grade orchestration platform for semiconductor suppliers.
A fab-impacting escalation is a business-critical event: yield, revenue, customer trust, and operational credibility are at stake. Affected material may still be moving as containment, shipment, qualification, and customer decisions remain live. Yet the response is fragmented across inboxes, spreadsheets, quality records, and engineering tools. Teams reconstruct exposure, reconcile facts, and chase owners while the impact can deepen.
FabQED orchestrates the response, not just the case file. One governed record connects the signal and affected scope to containment, investigation, evidence, accountable actions, reviewed customer updates, and signed closure. Teams can see what is known, what remains open, and who owns the next decision. Built for semiconductor supplier workflows across CMP, patterning, and advanced packaging, with AI assisting while people retain decision and communication authority.
Use the contact page to share the relevant context.
Contact us