Shahmurad Orujov
Baku
Summary
AI researcher with a strong foundation in electrical engineering, specializing in computer vision, deep learning, and generative modeling. Transitioned from electronics engineering to AI/robotics through an MSc in Robotics and Artificial Intelligence at the University of Glasgow, following a dual-diploma undergraduate program between South Korea and Azerbaijan.
Skilled in PyTorch, computer vision (CNNs, ViTs, ResNet, EfficientNet), generative models (VAEs, diffusion models, Stable Diffusion), and explainable AI (EigenCAM). Experienced in developing interpretable machine learning systems for animal behavior analysis, medical imaging reconstruction, and image synthesis. Whether I'm building emotion classification models that prioritize ethical reasoning, implementing diffusion-based medical imaging solutions, or exploring how generative AI can create controllable content, I'm always driven by impact: How can we design intelligent systems that are not only accurate but transparent, ethical, and truly beneficial for real-world applications?
Overview
3
3
years of professional experience
1
1
Certification
Work History
Postgraduate Researcher
University of Glasgow
06.2025 - 09.2025
- Replicated and enhanced baseline canine emotion classification methodology, improving test accuracy from 0.6687 to 0.7562 for ResNet50 (+13%) and 0.6937 to 0.7250 for MobileNetV2 (+4.5%) through strategic data augmentation techniques.
- Evaluated modern architectures (EfficientNetV2-M, ConvNeXt-Base, DINOv2 ViT-S/14) with unified 3-stage training pipeline, achieving up to 0.8621 accuracy on combined multi-source dataset compared to 0.6937 baseline (+24% improvement).
- Engineered combined dataset merging 3 distinct sources (6,596 images across 4 emotion classes) to address class imbalance and improve model generalization, resulting in more robust cross-dataset performance.
- Implemented EigenCAM-based interpretability analysis across 6 architectures, revealing CNNs focus on dog-centric features (face/body) while ViT models exhibit background bias, establishing explainability as validity criterion for animal welfare applications.
- Established reproducible evaluation framework with class-balanced metrics, stratified splits, and saliency-based validation, prioritizing animal-centric reasoning over raw accuracy for ethical AI deployment in Animal-Computer Interaction.
Undergraduate Research Assistant
Inha University
12.2023 - 02.2024
- Researched and implemented advanced generative models including Variational Autoencoders (VAEs), β-VAEs, and vector quantization VAEs (VQ-VAE) for data generation and representation learning.
- Studied denoising diffusion probabilistic models (DDPM), diffusion restoration models (DDRM), and decomposed diffusion sampling (DDS) techniques for high-quality image synthesis.
- Implemented sparse-view computed tomography (SV-CT) and multi-coil magnetic resonance imaging (MRI) reconstruction models for medical imaging applications.
- Developed Stable Diffusion models for text-guided inpainting with controllable parameters for content generation and image editing.
- Researched score-based generative models and stochastic differential equations (SDE) for solving inverse problems in medical imaging.
Electronics Engineering Intern
AzSmart Azerbaijan
12.2022 - 02.2023
- Repaired POS terminal hardware and PCB components.
- Performed initial hardware and software setup for the sales stage.
Education
Master of Science - Robotics And Artificial Intelligence
University of Glasgow
Glasgow, United Kingdom09-2025
Bachelor of Science - Electronic Engineering
Inha University
Incheon, South Korea07-2024
Bachelor of Science - Electrical And Electronics Engineering
Baku Engineering University
Baku, Azerbaijan07-2024
Skills
Python - Expert
Machine Learning - Expert
PyTorch - Proficient
MATLAB - Proficient
TensorFlow - Experienced
C - Proficient
Git - Proficient
Data Analytics - Proficient
SQL - Beginner
Languages
Azerbaijani
Native or Bilingual
Russian
Native or Bilingual
English
Native or Bilingual
Turkish
Full Professional
Korean
Professional Working
Spanish
Elementary
Certification
- Deep Learning Specialization, DeepLearningAI
- Data Analytics, Google Career Certification
- The Complete Python Developer, ZTM Academy
- C++, TestDome
Projects
- Canine Emotion Classification with Explainable AI - MSc dissertation achieving 86% accuracy using modern CNNs and ViTs with EigenCAM interpretability analysis, prioritizing animal welfare over raw performance metrics.
- IntelliFireCore Autonomous Firefighting System - Developed real-time, event-driven C++ firmware on Raspberry Pi 5 with Qt5 UI, integrating flame/distance sensors for autonomous fire detection and response.
- Computer Vision Lighting Control System - Built gesture-controlled lighting system using OpenCV and Python with Tkinter GUI, voice control fallback, and Arduino Uno hardware integration.
- LEGO Mindstorms "Robot Wars" - Programmed autonomous combat robot using EV3 Classroom with multi-sensor integration for opponent tracking and attack sequences, reached semi-finals.
- "QARTAL" Sumo & Line-Tracing Robot - Engineered LEGO EV3 robot with attachable claw mechanism for dual-purpose sumo wrestling and accurate line following using custom LEGO Mindstorms programming blocks.
- Laser Security System - Designed and implemented a laser-based security system, presented at the VII International Scientific Conference of Young Researchers with published proceedings.
- NFL Player Trajectory Prediction - Implemented Peephole LSTM with physics-informed constraints and Kalman filtering for 30-timestep player position forecasting using 5-fold cross-validation on 2023 NFL tracking data for Kaggle competition.