I’ve been writing code for 25+ years. The last few I’ve spent on a single problem: getting AI out of demos and into production — and keeping it there.
As AI Architect at Payoneer, I own the AI platform strategy for a domain of ~600 engineers: architecture, model selection, and infrastructure. That means multi-agent systems on AWS Bedrock AgentCore and Azure AI Foundry, a GraphRAG knowledge base with ~2.5B tokens indexed, and the MCP/LLM gateways that keep agents compliant in a regulated fintech — DORA, the EU AI Act, CPS 230, audit trails, and the kill switch.
On the model side, I work with fine-tuning — LoRA, QLoRA, DPO — on Azure AI Foundry and open-source models. Not because it’s always the answer, but because knowing when it is is half the job.
I’m also co-author of Payoneer’s AI-SDLC framework: redesigning the whole development cycle around AI agents, with humans in the loop where it matters. Building the platform is the easy part — getting 600 engineers to change how they work is the real one.