I speak about the unglamorous half of applied AI: what breaks when a language model leaves the demo, how you evaluate a pipeline you cannot fully inspect, and what inference actually costs.
Talks are drawn from systems I have shipped rather than from slideware, which usually makes the question-and-answer part the best bit.
Lviv AI Day · 2026 · Lviv, Ukraine
What actually breaks when a large language model moves from a demo into production: hallucination risk, evaluating multi-stage pipelines, and the cost of inference. Drawn from a screening pipeline running on local models inside a customer's own security perimeter.
SKELAR Backend Meetup · 2025 · Kyiv, Ukraine
Turning thousands of recorded lessons into something a product team can actually act on: the engineering decisions behind analysing them at that volume, and how the output is kept trustworthy enough to base product decisions on.
AWS User Group Kyiv · 2023 · Online
A snap talk at AWS Notes. Chapter 1. IoT about eSvitlo, which tracks whether there is electricity at a given address during the blackouts and tells people over Telegram and Viber. Ten minutes on the AWS side of it — Lambda, DynamoDB and SNS — and on the local hardware it reads: Wi-Fi routers and a Raspberry Pi.
If you're organising a conference or a meetup and this is useful to your audience, email me.
Email me