Building a multimodal, multilingual AI Agent

As part of an EU-funded innovation project, I worked as a Senior AI Engineer for a large telecommunications and IT service provider with over 2,000 employees. My task was to develop a multimodal, multilingual AI assistant. I was responsible for designing and implementing key AI components—including the actual AI Agent, as well as the individual components that it orchestrates - LLM with tool-calling abilities, Speech-to-Text, Text-to-Speech — as scalable microservices. The system was built with low latency in mind to support real-time communication, leveraging WebRTC for efficient audio and video streaming. The project also emphasized robust MLOps, CI/CD pipelines, and cloud infrastructure.