Data Scientist - Predictive Maintenance
Tractian Technologies, Brazil - Remote
Overview
Data Science at Tractian The Data Science team at Tractian builds machine learning systems that turn industrial data into reliable, actionable intelligence. Working with large-scale sensor data and complex time-series signals, we develop models that help industrial companies understand equipment behavior, detect anomalies, anticipate failures, and make better maintenance decisions. Our focus goes beyond model development. We build and evolve production-grade ML solutions, bridging advanced modeling techniques and robust software systems.
This includes designing scalable inference pipelines, monitoring model performance in real-world environments, addressing data and concept drift, and ensuring our models remain reliable as machine behavior and operating conditions change. What you'll do As a Data Scientist focused on Time Series and Predictive Maintenance, you will develop, deploy, and continuously improve machine learning systems that operate on real-world industrial data.
You will work across the entire ML lifecycle, from exploring new modeling approaches and validating hypotheses to building production-ready inference systems and monitoring their behavior at scale. You will tackle challenges involving non-stationary time series, heterogeneous equipment, noisy sensor data, distribution shifts, and the operational constraints of deploying ML in industrial environments. You will collaborate closely with Data Scientists, Software Engineers, and laboratory teams to turn research and experimentation into reliable products that deliver measurable value to our customers.
Responsibilities Develop time-series ML models: Design and implement machine learning algorithms for anomaly detection, forecasting, behavioral modeling, and condition monitoring using large-scale industrial sensor data. Productionize machine learning solutions: Translate experimental models into reliable, maintainable, and scalable production systems, considering inference latency, computational efficiency, memory usage, and operational costs.
Build model monitoring and evaluation systems: Design metrics, diagnostics, and monitoring mechanisms to track data quality, feature distributions, anomaly scores, prediction stability, and model drift and performance over time. Design robust modeling strategies: Investigate approaches that balance predictive performance, generalization, adaptability, interpretability, and maintenance costs under changing operating conditions. Engineer time-series representations: Develop feature extraction, signal processing, and representation learning techniques to capture meaningful temporal and physical patterns from industrial sensor data.
Run rigorous experiments: Formulate hypotheses, design controlled experiments, benchmark alternative approaches, and evaluate models using appropriate temporal validation strategies and realistic deployment scenarios. Optimize ML pipelines: Improve data processing, feature computation, model serving, and inference workflows to support high-throughput, low-latency, and resource-efficient applications. Bridge research and production: Evaluate state-of-the-art methods and translate promising techniques into practical, testable, and production-ready solutions.