I architect enterprise AI systems the way I'd architect a security perimeter: assume nothing, verify everything.
I build the governance layer that lets autonomous systems ship inside regulated enterprises — reusable platform architecture that domain teams inherit rather than rebuild. Currently AI Engineer Advisor at NTT DATA, leading AI modernization architecture across the Syntphony product ecosystem. Previously designed the Zero Trust and AI governance architecture behind a national-scale insurance data platform: work that became a registered design.
The systems I design tend to have the same shape: something autonomous needs access to something sensitive, and the architecture has to make that safe, explainable and operable. These are the concerns I actually own — the tools underneath them change.
Treating AI agents as first-class identities rather than service accounts borrowing a user's permissions — and passing real user identity through to the data layer instead of filtering after the fact.
Entra ID · OAuth2/OIDC · ZTNA · workload identity · RLS passthrough
Zero Trust structured properly: policy administration, decision, enforcement and information points as separate concerns, with continuous trust scoring instead of a binary allow/deny.
NIST 800-207 · PAP/PDP/PEP/PIP · adaptive trust scoring
Multi-agent topologies where domain specialists beat one monolithic agent, with standard tool protocols so capability is reusable across teams rather than rebuilt per engagement.
MCP · FastMCP · A2A · CrewAI · LangGraph · Semantic Kernel
Deciding what must stay live versus what can be indexed, and making every answer cite its source — including domain-aware parsing where generic chunking destroys retrieval quality.
Azure AI Search · hybrid search · semantic models · custom parsers
The layer above access control: per-agent rate limits, action-scope enforcement, full decision audit trails, and human escalation above a risk threshold.
OpenTelemetry · per-agent audit · human-in-the-loop gateways
Setting architecture direction for a team of engineers and data scientists — architecture reviews, decision records, and the trade-off conversations that precede them.
architecture governance · mentoring · stakeholder alignment
I arrived at enterprise AI architecture through security rather than the other way round, and that order is the whole point.
Architecture and technical direction across the Syntphony product ecosystem — agentic platforms on MCP and A2A, enterprise retrieval, legacy modernization.
Sole architect of the AI Trust Fabric: NIST 800-207 extended with continuous trust scoring and an AI governance layer, for a world where the thing requesting your data is an autonomous agent. Became a registered design.
Zero Trust segmentation and a SASE migration across 5,000+ users, replacing perimeter trust with identity-centric control. The grounding that everything after it depends on.
A national insurance data ecosystem aggregates sensitive records from every insurance company in the country. Perimeter security doesn’t address insider threat or an AI-era …
Mainframe modernization normally starts with senior COBOL engineers spending weeks manually reading code to reconstruct business logic and map dependencies — slow, expensive, and …
Platform architectures for AI describe how a system does more. Almost none describe what it is permitted to do. Here is the second half — a reference architecture for the control plane of an agent …
Enterprise agent systems fail at the disposition boundary, not the reasoning boundary. What an agent is permitted to do with its own output is an architectural decision, and it deserves a ladder …
Beyond Listening started as a music series. It turned out to be the name for how I approach most things — get past the surface of something, work out what is actually happening underneath, then build it or explain it. The architecture above is the same habit, pointed at systems instead of songs.
Latin styles doing narrative work — how bossa nova's unconventional chords lift a scene, and why those chords are the ones that fit.
Europe in 2019 with a guitar and no plan, Vietnam on Bourdain's advice, and a long monsoon drive through south India.
I'm open to Principal and Platform Architect conversations, particularly where autonomy, security and governance meet. The fastest route is email.