CV

Data Scientist

Baly.iq (Rocket Internet)

The #1 super app in Iraq.

• Engineered large-scale data processing pipelines using Python and Airflow to curate and clean unstructured datasets, directly enabling downstream ML training.
• Built reproducible Airflow pipelines and Python ETL modules to process unstructured and semi-structured fraud data, reducing manual work by 30%.
• Analyzed customer and transaction journey data to map data sources and improve anomaly detection accuracy.
• Tuned database systems using Percona PMM to speed up query response by 20%, ensuring 99.8% HA.
• Deployed containerized services with Docker and Ansible to deliver key business features faster when the tech team’s backlog delayed development; took ownership of deployment to keep operations on schedule.

Fraud Data Analyst

Snapp! (Rocket Internet)

The #1 super app and ride-hailing app in Iran.

• Investigated raw logs, relational tables, and metadata sources to design SQL and Python detection rules, identifying anomalies 40% faster.
• Collaborated on dataset curation for foundation model training by structuring raw logs into high-quality, ML-ready formats.
• Built tagging logic for fraud categories and structured irregular transaction flows into ML-ready datasets.
• Took initiative to learn automation and containerization, laying the groundwork for more scalable fraud-detection tools later used in production.

Project Lead / ML Engineer

Algorithmic Trading ML Infrastructure

Architected an end-to-end machine learning pipeline for algorithmic trading.

• Managed a 6-person engineering team using Kanban/Jira methodologies.
• Developed custom technical indicators for MetaTrader 5 integration.

MLOps Engineer

ML-IncomeInsightHub

Built a comprehensive machine learning dashboard prioritizing interpretability.

• Implemented SHAP value visualizations for feature importance.
• Engineered the backend using FastAPI, established secure geospatial routing with JWT authentication, and containerized the deployment via Docker.

Master of Science

Universiteit Van Amsterdam (UvA)

Data Science

• Thesis: Evaluating the Muon Optimizer vs. AdamW for High-Dimensional LLM Adaptation via LoRA.

MBA

University of Tehran

Business Administration

• Focus on strategic management and quantitative analysis.

BSc

Amirkabir University of Technology

Industrial Engineering

Expert Systems with Applications (ESWA)

Developed a Wasserstein GAN and dual autoencoder framework to solve severe class imbalance in financial datasets. Achieved near-perfect detection accuracy (0.9999 AUC) while maintaining extremely low false-positive rates.

Academic Project

Processed and analyzed a dataset of over 67,000 papers to extract trends in NLP research, authorships, and institutional impact.