IAȘI AI

Every AI demo seems to end with a model generating the final answer. But in high-stakes domains such as compliance, finance, and healthcare, a fluent response is not enough. Teams need outcomes they can reproduce, audit, explain, and defend months later.

Andrei's talk presents a different approach: use AI as an orchestrator of deterministic tools and well-defined workflows, rather than as the final decision-maker. Drawing on practical experience building AI capabilities for a KYC/AML compliance platform, we’ll explore architectural patterns that make AI systems more reliable, traceable, and trustworthy.

You’ll see why limiting the model’s role can increase, not reduce, its value, and how this shift gives engineers more meaningful problems to solve: designing robust tools, validation layers, decision trails, and systems that work beyond the demo.

Meet the speaker

Andrei Purdila Lead AI Engineer at spektr

I have spent the better part of a decade working at the intersection of big data and AI — six years at Amazon, two at Microsoft, and now as Lead AI Engineer at spektr, where I build AI into a compliance platform used by financial institutions for KYC and AML. That work sits in one of the least forgiving environments for AI: every decision needs to be reproducible, auditable, and defensible after the fact. It has shaped a strong point of view on what it actually takes to make AI systems trustworthy in production.