Experience

In the BitePal app team, I own AI features end-to-end — fine-tuning and evaluating VLMs for food recognition, portion-size estimation, and nutrient prediction, along with the synthetic-data pipelines behind them.

(RAI for Ukraine program) at the Center for Responsible AI at New York University, working on inference-time reuse and composition of pre-trained RL policies — when fixed policy libraries can be reused without additional training — under the supervision of Sihem Amer-Yahia and Noha Ibrahim.

Paper: Vitenko I., Ibrahim N., Amer-Yahia S. “When Can Pre-Trained Policies Be Reused? Inference-Time Policy Composition from Fixed Policy Libraries”. CIKM 2026, Short Paper Track. Extended version: arXiv:2604.20174.

Built multi-agent applications for clients in compliance, cybersecurity, and immigration, on top of a general-purpose agent framework using reinforcement learning from AI feedback (RLAIF). Notable builds: a VLM-powered compliance RAG system and an LLM+Diffusion storybook generator.

Built an automation-first solution for business workflows driven by LLM agents, and improved media-search accuracy with classical NLP and multi-speaker diarization.