I am a Master’s student in Computer Science & Engineering at UNIST and a researcher at the Interactive Multimodal ML Lab. My passion lies in exploring the frontiers of research and develop cutting-edge algorithms in the area of multimodal learning, particularly in the integration of vision and language. I am also interested in advancing the understanding and transparency of AI models. By bridging the gap between AI and human understanding, I aim to build more trustworthy and ethical AI systems that can be effectively utilized across various domains.

Besides my research interests, I have a broad interest in fintech, data science and the design / deployment of ML systems. I find great fascination in the practical application of machine learning techniques and their potential to solve real-world problems.

I always seek opportunities to expand my knowledge, stay updated with the latest advancements, and participate in meaningful projects that have a positive impact on society. I am enthusiastic about connecting and collaborating with fellow researchers, professionals, and organizations who share a similar vision.

• Programming Languages: Python, C++, JavaScript, R
• Frameworks & Tools: Pytorch, HuggingFace, LangChain, LLaMAFactory, Unsloth, DDP, DeepSpeed,
  FastAPI, Numpy, Pandas, OpenCV, Scikit-learn, Git, Docker, Latex
• AI/ML Expertise:
  • LLMs: LoRA/QLoRA, RLHF/DPO Fine-tuning, Retrieval-Augmented Generation (RAG) Integration, Quantization, Multimodal LLM Adaptation, Synthetic Data Creation, Prompt Engineering
  • Computer Vision: VLMs, CLIP, Image Classification (ResNet, ViT), Object Detection (YOLO, RT-DETR),
    Semantic Segmentation (U-Net, SAM), 3D Reconstruction (Gaussian Splatting)
  • Traditional ML: Linear/Logistic Regression, Support Vector Machines (SVM), XGBoost, Principal Component Analysis (PCA), K-means Clustering, Convolutional Neural Networks (CNN), RNN/LSTM

• RingFormer: Constructed a parameter-efficient and recurrent version of transformer models, named RingFormer, which maintains strong performance with significantly fewer parameters compared to the original transformer models across translation and image-classification tasks.
• Uzbek LLMs: Developed strong open-source LLMs for low-resource Uzbek language through language-specific continual pretraining and instruction-tuning with synthetic data. Released models on HuggingFace for broad accessibility.
• FAQ Chatbot: Built a RAG-based chatbot for NAVER Smart Store FAQs using FastAPI, Chroma, and OpenAI embeddings; integrated multi-turn memory, real-time streaming, and semantic similarity thresholding with prompt-level guardrails for handling out-of-scope and insufficient-context cases, ensuring safe and accurate responses.
• Guessing Game with Robot Arm: Designed a robot arm system and interactive UI for a guessing game using vision-language models (InternVL, YOLO-World) as part of a team, with TTS/STT integrated to enhance user experience. The robot interprets user clues, identifies objects, and responds via text and speech.
• 4D Instruct-GS2GS: Extended semantic editing to dynamic 3D scenes using an iterative dataset update and efficient 4D-GS rendering method for consistent editing of Gaussian splatting scenes via text instructions.
• Forecasting Transaction Fees on the Ethereum Blockchain Network: Modeled Ethereum gas fee dynamics using SARIMA and LSTM models, identifying daily/weekly usage patterns and achieving better test performance by capturing long-term dependencies.
• Portfolio Allocation Stability with CorrGAN: Investigated the robustness of traditional and ML-based portfolio allocation methods using simulated return data with random shocks and correlation matrices generated via CorrGAN model.
• General ML Applications: Developed various ML applications, including house price prediction using Lasso and GradientBoost methods, income classification with Logistic Regression, K-Nearest Neighbor and Random Forest, and healthcare expenditure modeling using regression analysis to examine the statistical significance of factors such as obesity influencing medical costs among the elderly.
• Data Structures for All Tastes: Implemented a variety of data structures from scratch in C++, ranging from simple ones like Linked Lists to more advanced types such as Balanced Binary Search Trees.

• Lotte Scholarship for International Master's Students: Awarded by the Lotte Scholarship Foundation for students with exceptional research potential in Korea, providing KRW 12,000,000 per year for the duration of graduate studies.
• Korean Government Scholarship for Graduate Studies: Comprehensive scholarship including a tuition fee waiver and a stipend of KRW 9,600,000 per year throughout graduate studies at UNIST.
• UNIST Global Dream Scholarship: Covered tuition fees, health insurance, and living expenses during undergraduate studies at UNIST.

• Machine Learning Engineering for Production Specialization – DeepLearning.AI specialization covering scalable production pipelines, ML model deployment, and MLOps best practices.
• Deep Learning Specialization – Taught by Andrew Ng, focusing on neural networks, convolutional networks, and sequence models for real-world AI tasks.
• Applied Data Science with Python Specialization – University of Michigan course series focusing on Python-based data analysis, including Pandas, Matplotlib, and machine learning fundamentals.
• Statistical Learning – Stanford Online course covering supervised and unsupervised learning, linear regression, and classification methods.
• Machine Learning – Classic Coursera course by Andrew Ng, covering algorithms, supervised/unsupervised learning, and neural networks.
• Mathematics for Machine Learning Specialization – Imperial College London specialization focusing on linear algebra, calculus, and statistics for machine learning foundations.

I am native in 🇺🇿 Uzbek, fluent in 🇷🇺 Russian and 🇬🇧 English, intermediate in 🇰🇷 Korean!