
The Missing Text Layer: Why Speech-to-Speech Agents Fail Enterprise Requirements
Native speech to speech demos beautifully, then fails enterprise needs. Why the text layer, where guardrails and audit live, still wins regulated voice.
Technical Project Manager
Yuri builds conversational AI agents for financial institutions and gets them live. That covers prompt engineering, data pipelines and rollouts in many languages across the Middle East, Africa and Europe.
Yuri is a Technical Project Manager at Blits.ai. He builds and delivers the conversational AI agents that enterprises actually put in front of their customers. The job mixes prompt engineering, data architecture and getting things deployed: he takes a client's business requirements and turns them into an agent that holds up in production, and not only in a demo room.
He builds the RAG pipelines, sets up the evaluation frameworks and runs the test cycles. Those decide whether an agent gives a customer the right answer, or the wrong one with total confidence.
"The data pipeline is the product as much as the conversation is. If you haven't tested how your agent breaks, you don't know how it works."
Language is a big chunk of the work. Blits.ai supports over 100 languages and Yuri keeps testing them. When a bot talks to customers in their own language, the way they really speak it, it feels like a helper. When it falls back on generic English it feels like a foreign system that doesn't get their world. Saudi dialect Arabic, South African vernacular: if the cultural and linguistic register is off, you have a translation layer and not a working agent.
"The difference between a good demo and a good product is about a thousand hours of evaluation nobody sees." (Yuri Ihnatov, Technical Project Manager)
Notes from the field on agentic AI, evaluation and shipping conversational agents in financial services

Native speech to speech demos beautifully, then fails enterprise needs. Why the text layer, where guardrails and audit live, still wins regulated voice.

Five takeaways from Mastercard's summit in Johannesburg: the technology is no longer the bottleneck, governance, trust and context are.