host
guest · Head of NLP at T-Bank and an ML and GenAI engineering leader
Artem Bondar is Head of NLP at T-Bank. He has spent more than a decade at the intersection of software engineering and machine learning, building ML and GenAI products and automating large business processes.
Alexander Polomodov and T-Bank Head of NLP Artem Bondar examine GenAI outside software development. Four cases show that value comes not from placing a model inside an old process, but from redesigning work around a measurable outcome, short feedback, and accountability.
Code supports models with compilers, linters, and tests. Operational feedback is sparse, so autonomy needs a verifiable environment. In one support experiment, a procedure explained only about 20% of decisions. Before automation, teams must recover stages, exceptions, and criteria for correct behavior.
In support, resolution and escalation matter; in accounting, calculations stay in code while models handle documents. In marketing, AI works better as an early reviewer than a copy factory. In building design, it explores options while calculations verify strength. Costs include review, corrections, and penalties.
Bondar recommends owning the vertical slice to customer outcome: faster work may only move the queue, and manual control can erase the gain. Teams ask which metric moves, how defects are found, and whether control costs more. Human in the loop is no guarantee; a named owner must treat agent failures as their own and change rules, context, or escalation.