The butler-style governance flagship — beyond side-channel monitoring, it actively steers and influences LLM/agent behaviour: goal anchoring, challenge escalation, hypergraph fleet governance. It embeds the HoloGovernor monitoring engine (transaction-level guarding, tamper-evident audit chain, governance reports), mapping to the Agent chapter of the AI Safety Governance Framework 3.0. Driving analogy: HoloManager is the steering wheel; HoloGovernor is the speed limiter + dash cam. Three related patent applications were accepted by CNIPA (CN202611615191.1 / CN202611615185.6 / CN202611615186.0).
Launched globally at the 2026 World Young Scientists "AI+" Dialogue in Macao. HoloAuditor audits AI output against geometric manifolds (0.82 ms interception); HoloBlocker guards the physical-world boundary with topological fusing. The launch also presented the "6.0% semantic-ranking collapse" finding — bigger models do not mean steadier reasoning.
Translates the holistic experience of TCM pattern differentiation into tensor manifolds and hypergraph audit, making TCM data computable, trustworthy and traceable — the foundation for TCM data-element circulation and cross-border regulatory mutual recognition. Theory led by Chair Prof. Ng Kwok-Po of HKBU (SIAM Fellow), pioneer of Block Tensor Decomposition (BTD).
Systematically exposes the reward-hacking and runaway risks of "survival motives" in embodied AI (self-funding agents), and proposes a three-layer architecture — safety manifold, topological distance, millisecond fusing: behaviour off the manifold is physically cut off. Better a blown fuse than a disaster; we advocate physical fusing as a national safety standard.
Replaces black-box classifiers with "frozen representations + topological inference": a minimal-parameter, zero-shot LLM guardrail with alert mode — an auditable, self-hostable content-safety gate for open models.
A cross-family, cross-metric dual-track empirical audit of intermediate-representation leakage in split learning / split inference: turning "leakage risk" into measurable, comparable, fresh numbers, with AMBER/RED-graded reports supporting data-export decisions.
Building the "Digital Great Wall" for adolescent physique data and a K12 AI-driven research network coupling data production with talent innovation — the trusted data foundation for Hong Kong's Digital Education Blueprint and the national "Data Element ×" strategy.