Atharva Hatekar

Developed a production-ready anomaly detection framework for multivariate industrial time-series data, evaluating Static-PCA, Adaptive-PCA, and a Graph Deviation Network (GNN) to improve predictive maintenance and process monitoring. Built reusable Python data preprocessing and feature engineering pipelines for 5+ years of sensor data, reducing data preparation time by 30% and enabling reproducible model development. Engineered Dockerized Python services with gRPC, CI/CD, Azure, Git, and comprehensive unit/integration tests to improve reliability and maintainability. Designed and deployed a root-cause diagnostic Grafana dashboard for real-time equipment monitoring to enable proactive anomaly detection by process engineers.

Built a real-time process monitoring system for 100+ process sensors. Benchmarked statistical outlier detectors (Isolation Forest, LOF) alongside time-series forecasting models (ARIMA, SARIMA, Prophet, LSTM) for drift detection and predictive maintenance on live plant data. Automated preprocessing, feature engineering, and exploratory analysis pipelines for high-dimensional time-series data and collaborated with process engineers to validate models using domain expertise.

Connected ACT process control software with gRPC and built a Grafana KPI dashboard combining domain-informed process health metrics with linear regression on sensor trends to detect early signs of degradation for predictive maintenance. Collaborated with cross-functional teams to diagnose data-quality issues, define data-wrangling standards, and establish KPIs and validation criteria for downstream analytics.