CO-FOUNDER & CTO · QUANTUMIZE AI
Agents that hold in production.
I design and build the architecture behind our products: agents that know your business, show their sources, and leave the decisions to you.
THE REAL RISK
Adopting an AI platform means handing over the one asset you're least willing to lose control of.
Adopting an AI platform means handing over the one asset you're least willing to lose control of: everything your company knows.
Context you can't audit
Documents, CRM records, and tribal knowledge get vacuumed into a vector store nobody can inspect. When an output is wrong, no one can say why, or which source it came from.
Answers with no chain of custody
An agent tells your customer something. Which document did that come from, and how current is it? When nobody can answer that, the confidence in the answer is doing all the work and you're the one carrying the risk.
Control you never decided to give up
Your context ends up in a training corpus, in another jurisdiction, under terms nobody on your side read. An agent commits to something no human approved. Nobody chose any of this; it accumulated by default.
The guardrails around an agent decide how much you can trust it.
WHAT I OWN
Five things I put my name on.
Agents built around the work
An agent earns its place by taking a real task off someone's plate. Argo Sales turns an hour of prospect prep into a five-minute action plan covering the opening angle, competitive positioning, and the objections your seller will actually hear. The design starts from the work your team already does, not from a chat box.
Context Engine
This is what makes those agents useful. Your company's knowledge is modeled as a governed graph rather than dumped in as documents, with entities, relationships, and provenance all explicit, so every answer traces back to a source and a date.
Tenant separation
Your data is separated from every other customer's by design, with the isolation model matched to your risk profile. The boundary is enforced in the data layer, not just in application code.
Grounding & evaluation
We measure retrieval quality instead of assuming it. Releases are gated on retrieval accuracy before an agent gets anywhere near a customer conversation.
Operability
Tracing on every agent run, error monitoring, staged deploys, rollback. When something misbehaves, we can show you the exact run: prompt, retrieved context, model, latency.
WHY IT'S BUILT THIS WAY
None of this is theoretical caution. It comes from years of platform work inside large organizations.
None of this is theoretical caution. It comes from years of platform work inside large organizations, and that work changes your instincts.
CAE: data & governance platform, in production
I led design and delivery of the proprietary catalog, metadata-driven acquisition and ingestion, automated profiling and quality controls, Kubernetes orchestration, and SOC 2-aligned consumption on Databricks. Everything there was shaped by one constraint: governance had to accelerate product teams instead of blocking them.
MTY Food Group: platform + apps
Platform Architect for a multi-banner data and AI platform and the applications consuming it. Available ≠ actionable.
CAE Rise: microservices & near-real-time
Python and C#/.NET services, near-real-time analytics, Azure infrastructure. Working that close to operations is where platform decisions started to make sense to me.
In a regulated enterprise you find out fast that a demo and a platform are not the same thing.
The CAE and MTY work was delivered in prior operating roles, before Quantumize AI.
→ Technical detail on both buildsLonger notes on the problems above: context engine design, multi-tenant isolation, retrieval evaluation, and what actually breaks when agents meet enterprise data.
WHAT MY TEAM AND I ARE BUILDING
Three products, one engineering standard.
Three products, one engineering standard. All three rest on the same two internal standards (how an agent is structured, and how a domain's knowledge becomes context an agent can reason over), share one design system, and follow one rule: nothing ships without passing its evaluation set. The stacks diverge where the problems diverge, but the discipline stays put.
One is in enterprise pilots. The other two are still being built, and they're labelled that way below.
in enterprise pilots
Argo Sales
One to two hours of prospect preparation turned into a five-minute action plan covering the opening angle, competitive positioning, and objection handling. Ten-plus users are on it in active enterprise pilots.
in development
Argo Marketing
An agentic marketing operating system for smaller brands. It interviews the founder to build a structured picture of the business, then runs a weekly loop: audit how the brand surfaces in search and in AI answers, rank what's worth fixing, produce the content. Prioritization is deterministic, and corrections wait for a human to approve them.
in development
Ludo
An AI football-law assistant for licensed agents, club executives, and sports lawyers, grounded in the FIFA regulatory and dispute-resolution corpus. Every legal claim traces to a retrieved source. If a citation comes out of model memory, we treat it as a defect.
By the third product you find out whether the first one was architected or improvised.
Everything here is built with the Quantumize engineering team. I don't take work outside it.
Need help making AI actually land in your organization?
Bring the hard questions about architecture, data handling, and what happens when an agent gets it wrong. I answer those myself, without a handoff to a sales engineer.