At Sber’s Center for Practical AI, I prototype autonomous LLM agents from recent research and translate research prototypes into production-ready code. I built a custom LangGraph runtime with long-horizon planning, self-correction, and tool use.
I test planning architectures through A/B experiments on synthetic benchmarks and real-world scenarios, and explore agent memory using vector stores, graph databases, and scratchpad notes. I also participated in RLAIF/RLHF dataset construction and investigated agent failure modes.
My research has included fine-tuning large language models using RLHF at T-Bank Research and quantum machine learning research at Terra Quantum. In a VLM project, I fine-tuned pretrained models with LoRA and QLoRA and benchmarked BLIP-2, InstructBLIP, and LLaVA.
Earlier, I developed Flutter applications and C# backend services, including payment systems and Web3 smart-contract integrations. I also won a Bronze Medal in the 2026 NVIDIA Nemotron Model Reasoning Challenge.
