At EPAM Systems, I design and maintain production-grade Agentic RAG and LLM-powered applications for enterprise document workflows. I build backend APIs and custom AI-agent components for platforms serving 70,000+ users.
I architect Agentic RAG workflows with Azure AI Search, routing logic, memory integration, and grounded response generation. I also develop document-intelligence and ETL pipelines, and refine AI performance through retrieval testing and failure analysis.
At CyberNeuron, I worked on WoofSense, a real-time dog-vocalization emotion recognition system. I trained classification models that achieved 96% accuracy and applied knowledge distillation to achieve up to 10× faster training while preserving model accuracy and inference quality.
As a Computer Vision Researcher at FAST Foundation, I developed a multi-branch semantic segmentation architecture for aerial imagery and benchmarked it against 12 state-of-the-art models, achieving SOTA performance. I’ve also taught machine learning at Yerevan State University and Plekhanov Russian University of Economics (Yerevan Branch).
