Andrii Diakun

I design, research, and build AI systems — from neural network architectures and multilingual language models to retrieval-augmented generation and agentic software.

Since 2021, my work has evolved from rule-based approaches and early neural architectures to custom attention layers, EVA-I, a multilingual language model developed with PyTorch, and production AI systems deployed in real business environments.

As the founder and lead developer of Revelint, I work across the full AI development lifecycle: model architecture, data pipelines, retrieval systems, application engineering, deployment, and product development.

Research interests

• Multilingual language models and neural architectures
• LLM pretraining data curation, deduplication, and evaluation
• Retrieval-augmented generation and knowledge-grounded AI
• Agentic systems and AI-assisted software engineering
• Resource-efficient AI and adaptive learning technologies

My approach combines hands-on experimentation with production engineering. I am particularly interested in making AI systems measurable, reproducible, and useful beyond controlled demonstrations.

I am interested in research collaborations, joint AI R&D initiatives, and academic partnerships across Europe, particularly in multilingual NLP, applied machine learning, and AI for education.

GitHub: github.com/AndriiDiakun