Selected work
Case studies
Anonymised engagements across cybersecurity and AI. Every case follows our provenance-first methodology.
Agentic LLM pipeline for regulatory due-diligence
A multi-agent system that executes Zhianrui’s provenance-first methodology in code — sourcing, tier classification, contradiction detection, confidence scoring.
Production eval harness for a safety-critical LLM application
A continuously-running evaluation suite that catches the failures that matter — not benchmark scores, but the specific regressions a regulated domain cannot tolerate.
Provenance-aware RAG for legacy engineering documentation
A retrieval-augmented generation architecture for technical documents in regulated environments — every generated answer maps back to its originating document, page, and revision.
Synthetic persona modelling for public-sector decision support
AI-generated stakeholder personas grounded in survey data, demographic registries, and policy-position datasets — used to stress-test policy proposals before public consultation.
Three-zone authentication for legacy industrial control systems
Hardware root of trust retrofit for OT/ICS environments with IEC 62443 compliance pathway.
Biosecurity lifecycle monitoring with ML-driven anomaly detection
End-to-end provenance chain for biological materials using AI-augmented screening.
Post-quantum key exchange for sovereign messaging infrastructure
NIST-aligned PQC migration for a European government communications platform.
AI-native generative design for structural engineering
Topology-optimised component design using constrained generative AI within FEA validation loops.
Zero-trust architecture for a whistleblower reporting platform
End-to-end encrypted submission pipeline with jurisdictional data-residency guarantees.