AI agents · AI governance · Enterprise platforms
Built and shipped multiple products from the ground up, and helped architect an enterprise compliance platform used by Fortune 500 customers worldwide.
I build production agentic AI systems end-to-end — multi-agent orchestration, MCP servers, RAG, and graded LLM evaluation harnesses.
I treat AI agents as untrusted actors, not trusted infrastructure: agent boundaries, audit trails, alignment with emerging AI regulations, and human approval gates before anything irreversible.
Led a move from an always-on monolith to on-demand Kubernetes.
I build voice and multi-channel agents with evaluation harnesses, regression baselines, structured-output checks, and human approval gates.
I'm a senior engineering leader. I've led globally distributed teams and helped architect, from its earliest stages, an enterprise compliance platform used by Fortune 500 customers worldwide. My current focus is AI in enterprise software: I've shipped enterprise AI-agent systems and defined AI strategy and governance for the GRC platform — agent boundaries, audit trails, and alignment with emerging AI regulations.
I grew from software engineer to Senior Director. This year I've shipped several enterprise AI-agent systems, and took one security agent from concept through architecture to working code myself.
On my own time, I build LLM systems end to end: a live voice agent on OpenAI's Realtime API, a multi-channel agent platform, MCP servers. Most of that effort goes into graded evaluation harnesses that run against live models with committed regression baselines, structured-output guardrails, and human approval gates before anything irreversible. I still write code; it's the best way I know to see where this technology actually is, rather than where the demos say it is.
I've led a globally distributed organization of 15 engineers across multiple teams, owning the full lifecycle of a compliance platform Fortune 500 companies run their audits on.
I care personally and challenge directly, with feedback early enough to act on and expected in both directions. Trust comes from being useful, not agreeable.
Ownership, curiosity, and a bias for action: carry a problem past your own component, go read the incident, the code, the regulation, and ship a reversible decision this week rather than a perfect one next quarter.
Across time zones, decisions need to survive without me in the room, so I rely on written acceptance criteria, one accountable owner per workstream, and roadmaps visible before they're committed. When priorities collide I cut scope in the open.
I've been a key interviewer across the broader organization and helped shape the hiring bar, so I treat team composition as a design problem rather than a headcount problem. I'd rather grow a senior engineer into a lead than backfill from outside, and I plan for that ahead of needing it.
I mentor early-career programmers and coach developers and tech leads on AI adoption. The goal is not more lines of code; it is turning unclear requirements into working systems and owning them in production, without giving up quality, security, or accountability.
On AI governance: I treat AI agents as untrusted actors, not trusted infrastructure. The same Zero-Trust principles that apply to network security apply to AI systems — least-privilege access, hard boundaries, behavioral monitoring, human checkpoints on high-stakes decisions, and well-defined kill switches.
AI changes how the work gets done. It doesn't change why the job exists.
I write about building AI systems, engineering leadership, and the controls enterprise software needs.
I argue for treating AI agents as untrusted actors rather than trusted infrastructure. The controls that matter — tool allow-lists, data boundary rules, human checkpoints, kill switches — only work when they're enforced outside the agent's own process. Guardrails an agent can talk itself out of aren't guardrails.
I think the engineers who thrive won't be the best coders — they'll be the best managers of AI agents. Directing an agent and growing a junior developer need the same things: clear acceptance criteria, real code review, honest task decomposition. The dividing line forming isn't technical versus non-technical. It's delegators versus non-delegators.
Hands-on lessons from building multi-agent architectures: prefer specialized agents over monolithic ones, reserve the model for work that genuinely requires understanding, and let deterministic code handle everything else. The AI is roughly 20% of the build — architecture, data flow, error handling, and testing are the other 80%.
I publish research-backed deep dives on retrieval-augmented generation, citing Stanford HAI and Google DeepMind, and I make the case for hybrid retrieval — BM25 plus dense — re-ranking, and plain honesty about what RAG can and cannot do.
More on LinkedIn.
Nexus by Yuval Noah Harari
When I'm not building software, you'll probably find me on the water. I'm an ASA-certified sailor — keelboat through bareboat charter, monohull and catamaran — and a few days of coastal cruising with no signal is still the best reset I've found.
An AI assistant trained on my full background answers in seconds — leadership style, how I run teams, what I'd build in your context. Recruiters can leave a message right in the chat, or reach me on LinkedIn.