I'm an Information Systems student at Carnegie Mellon University, concentrating in Data Science and Privacy & Security, with an AI for Finance certificate. My work centers on machine learning and computer vision, from building synthetic data pipelines for wildlife counting to real-time 3D object detection, and I enjoy branching out into blockchain and database systems along the way.
 | Built a geospatial ML pipeline to predict deforestation risk across 342,819 grid cells in the Brazilian Amazon, combining five open data sources. Final XGBoost model achieves AUC-ROC 0.990 and Recall 0.947. Tech Stack: Python, XGBoost, scikit-learn, geopandas, pandas View Repository |
 | Built a 6-class image classifier for aerial and side-view herd imagery across 9 experiments. Final model (ResNet50 + SVM) achieves 93.30% test accuracy on 5,870 labeled images spanning Buffalo, Elephant, Hoofed Grazers, Musk Ox, Wildebeest, and Zebra. Tech Stack: Python, ResNet50, scikit-learn, TensorFlow/Keras, OpenCV, Selenium View Repository |
 | Generated synthetic animal group datasets in Blender to improve object counting and detection. Evaluated YOLOv8 and CSRNet on fish, bird, and mammal datasets. Tech Stack: Blender, Python, YOLOv8, CSRNet, PyTorch View Repository |
 | Developed a full database design and implementation for a subscription and offer-redemption platform. Includes UML modeling, BCNF normalization, and analytical SQL queries. Tech Stack: PostgreSQL, SQL, Python View Repository |
 | Implemented a blockchain-based system for secure, immutable tracking of supply chain transactions using smart contracts to ensure data integrity. Tech Stack: Solidity, Blockchain View Repository |
Generating Synthetic Datasets to Train Deep Models for Counting in Large-Group Animal Imagery ICMV 2025 (Paris)| Best Oral Presentation, MotM: 1st Place in Extensive Research Advised by Gianni Di Caro, CMU Professor in Computer Science & AI departments Designed an end-to-end synthetic data generation pipeline (Python + Blender) producing 50,000+ annotated training images using the Boids flocking algorithm, and trained YOLO/CSRNet models achieving a 90% improvement in counting accuracy over benchmark models.
Community Detection in Dynamic Face-to-Face Interaction Networks HARP Journal, 2023 Advised by Dr. Maria Konte, Georgia Institute of Technology Applied a tailored Louvain community-detection algorithm to temporal face-to-face interaction networks, optimizing the resolution parameter (γ) for fine-grained community structures, with a full Python pipeline using NetworkX and Matplotlib.
Email: iibrohim@andrew.cmu.edu
LinkedIn: www.linkedin.com/in/iroda-ibrohimova-73098924a