LLM research · AI engineering · reproducible ML
Third-year Applied Artificial Intelligence student at Innopolis University (BSc 2024–2028).
Portfolio · Research CV · Engineering CV · AI Agents CV · Kaggle · Email · ORCID
I study LLM quality through held-out evaluation and build auditable AI services with RAG, human review, and reproducible model delivery.
- Modeling & retrieval: Python, PyTorch, TensorFlow, Hugging Face Transformers, scikit-learn, FAISS
- Research methods: held-out evaluation, leakage-free splits, bootstrap confidence intervals, reproducible experiments
- Applications & delivery: FastAPI, PostgreSQL, DVC, MLflow, ClearML, Apache Airflow, Docker, GitHub Actions
- SemanticSplat — led a four-person research team through a 150-query evaluation across five scenes, with ScanNet and Replica pilots.
- BudgetRoute-LLM — compared five Qwen2.5 routing strategies over 1,000 runs on 500 MMLU questions with held-out evaluation; built traceable RAG, abstention, and human review.
- Yacht Resistance MLOps Pipeline — automated DVC, MLflow, FastAPI, and Streamlit delivery with 10 CI tests; held-out R² 0.993 and RMSE 1.28 on five unseen hulls.
- Reliable Image Classification with Confidence Rejection — completed a four-person computer vision project on CIFAR-10 classification with confidence-based abstention; implemented a ResNet baseline, calibration and rejection analysis, and evaluation on blurred, noisy, and JPEG-compressed images.
