Niloofar Rahimi

Tehran, Iran | niloofar.rahimi@email.com | +98 912 000 0000 | linkedin.com/in/niloofar-rahimi | github.com/niloofar-rahimi

SUMMARY

Machine learning engineer with 5 years of experience building and operating ranking and recommendation systems at consumer scale. Owned models serving 8M+ daily sessions, cutting inference cost 35% while lifting conversion 12%. Deep production focus: model serving, A/B experimentation, and MLOps on Kubernetes.

SKILLS

• Languages: Python, SQL, Scala, Bash
• ML: PyTorch, XGBoost, scikit-learn, Transformers, learning-to-rank, recommender systems
• Data and Infrastructure: Spark, Airflow, Kafka, Docker, Kubernetes, MLflow, AWS
• Practices: A/B testing, CI/CD, model monitoring, feature stores, offline evaluation

EXPERIENCE

Senior Machine Learning Engineer — Digikala, Tehran | 2022 – Present

• Redesigned the product-ranking pipeline into a two-stage retrieval and re-ranking architecture, lifting add-to-cart rate 12% across 8M daily sessions.
• Cut inference cost 35% by distilling a transformer ranker into a gradient-boosted model that retained 98% of offline NDCG.
• Built a feature store consolidating 200+ signals, reducing new-model development time from six weeks to ten days.
• Mentored 3 junior engineers and introduced an offline evaluation framework adopted by four product teams.

Machine Learning Engineer — Snapp, Tehran | 2020 – 2022

• Developed demand-forecasting models covering 30 cities, reducing driver idle time 9% and improving ETA accuracy 18%.
• Shipped a fraud-detection service scoring 2M rides per day at sub-50ms latency, cutting the false-positive rate 40%.
• Automated retraining pipelines with Airflow, eliminating 15 hours per week of manual operations.

EDUCATION

M.Sc. Computer Engineering (Artificial Intelligence) — Sharif University of Technology | 2018 – 2020

Thesis: session-based recommendation with graph neural networks. GPA 18.4/20.

B.Sc. Computer Engineering — Amirkabir University of Technology | 2014 – 2018

PROJECTS

• Core contributor to an open-source learning-to-rank library (500+ GitHub stars); implemented listwise loss functions and evaluation utilities used in three published benchmarks.