Graduate Research Assistant · Nazarbayev University · Astana, Kazakhstan
M.Sc. Computer Science graduate and Graduate Research Assistant at Nazarbayev University, working on IoMT health monitoring and multimodal human activity recognition under Prof.
Adnan Yazici. My research centers on metaheuristic feature selection, sensor fusion, federated learning, and anomaly/emotion/fall detection for wearable and video-based health systems. I build end-to-end ML pipelines in PyTorch: from data preprocessing and feature engineering to deployment, and also have experience developing full-stack applications (FastAPI, React). Recently published in PeerJ Computer Science (Q1) on gender-aware speech emotion recognition, with a second manuscript on automated feature selection for HAR currently under journal review. Another survey paper on multimodal HAR has also been submitted and is currently under review.
Nazarbayev University
Jan 2025 — PresentMetaheuristic feature selection and multimodal HAR/health-monitoring research (PyTorch) under Prof. Adnan Yazici — 96–97% LOSO accuracy across datasets, one PeerJ Q1 publication, two manuscripts under review.
Nazarbayev University
Dec 2022 — Nov 2023Trained 3D CNN models on facial, body-landmark, and emotion features (PyTorch) to predict human intention. Reached 77% accuracy on a custom video dataset.
Nazarbayev University
Jan 2025 — Dec 2025TA for Intro to Computer Science (1st-year) and Artificial Intelligence (3rd-year) over two semesters: prepared practice materials, graded assignments/exams, proctored exams, and supported students through concept and exam reviews.
Nazarbayev University
2024 — 2026Coursework in Generative AI, Deep Learning, and Computer Vision; thesis research on metaheuristic feature selection for multimodal human activity recognition.
Nazarbayev University
2020 — 2024Foundational coursework in computer science; undergraduate research assistant work on 3D CNNs for human intention estimation.
Yessenov Data Lab
2025 — 20254-week intensive fellowship covering data science, computer vision, and LLMs; selected among the top ~10% of national applicants.
* A metaheuristic feature selection framework for multimodal HAR, benchmarking 14 algorithms on UTD-MHAD and CZU-MHAD. * Reached up to 96.99% (UTD-MHAD) and 97.45% (CZU-MHAD) LOSO accuracy, beating MI/RFE/Lasso baselines while cutting feature dimensionality by up to 80%. * 10–15% cross-dataset accuracy gains under simulated sensor failure and on unseen datasets. * Submitted for journal publication
Open project* Built a CNN-based SER pipeline with gender-based segmentation, training gender-specific models to reduce acoustic variability between speakers. * Applied feature selection on acoustic features (MFCCs) to improve efficiency + reduce dimensionality. * Tested on Emo-DB, EMOVO, and RAVDESS; gender-aware pipeline improved accuracy by 10% over the gender-agnostic baseline. * DOI: 10.7717/peerj-cs.4016
Open projectFull-stack platform (FastAPI backend, JS frontend, deployed on Vercel) delivering lessons and practice tests for 6th-grade students. Actively used in production by 75-80 students.
Open project* Built a single-label classification pipeline from the multi-label Pascal VOC 2012 detection dataset. * Trained/evaluated CNN, ResNet50, EfficientNet-B0, MobileNetV3, ShuffleNetV2, ConvNeXt-Tiny in PyTorch * Addressed class imbalance; reached 82.78% test accuracy with MobileNetV3 (Adam, head-only fine-tuning).
Open project* Multimodal HAR pipeline on the UP-FALL dataset (video + wearable sensors) * Transforming inputs into Summed Difference images, Gramian Angular Fields, and Recurrence Plots. * Reached 91% test accuracy with deep learning fusion models.
Open projectFine-tuned aMUSEd and Stable Diffusion 1.5 on the SciCap dataset for generating scientific figures Improved SSIM, FID, and CLIP scores by 15%.
Nazarbayev University, ACM-W · 2023
Achieved the first place at a team competition across three disciplines: competitive programming, maths, system design.
Nazarbayev University, ACM-W · 2023
Got absolute 2nd place in a competitive programming contest CodeW in the "Advanced" track
