Gagan Saini

Data Engineer I — CloudEQ Software India Pvt. Ltd.

Jul 2022 – Dec 2024 | India

• Designed a machine learning-based resource clustering algorithm in Python utilizing Scikit-learn, TensorFlow, and DBSCAN to detect utilization patterns for 500+ cloud resources, uncovering cost-saving patterns that led to a 25% decrease in client cloud spend
• Created automated ETL flows using Python and SQL for the collection, cleaning, and transformation of massive infrastructural telemetry from AWS, Azure, and GCP API endpoints — saving up to 40% time on data preparation for statistical analysis
• Created an anomaly detection system using Python applying threshold-based statistical models to live metrics data of 500+ monitored infrastructural elements, detecting performance degradation trends and decreasing incident response time by 30%
• Used feature engineering and clustering algorithms in Python for multi-source cloud utilization data to create clusters of similar infrastructures, providing structured analytics data for further analysis
• Implemented scalable data pipelines for ingestion of structured data using Microsoft Fabric Data Factory technology

Bayesian Synthetic Fraud Dataset Generator for Music Streaming

Built in collaboration with XYNQ for their ORIGIN fraud detection platform. An expert-elicited Bayesian Network with a hand-crafted DAG (10 nodes, 12 edges) generating 200,000 synthetic music upload records — grounded in published fraud cases including a federal prosecution involving $10M in misappropriated royalties.

• Two-phase generation pipeline embedding real-world fraud signals: audio hash reuse (4.6×), IP rotation (40%), and temporal bursting
• Validated with Spearman ρ, Cramér's V (Bergsma bias-corrected), and correlation ratio η — mean absolute deviation ≤ 0.01 across all key variables
• Confirmed statistically significant class separation (user agent 0.77, duration 0.45) suitable for downstream fraud classifier training
• Deterministic, dictionary-free phoneme-based text generation for reproducible artist names and track titles without hardcoded word lists

Tech: Python, Bayesian Networks, NumPy, Pandas, Statistical Validation