SFCL Tech Inc: Senior Machine Learning Engineer-Anomaly Detection (BESS). Job Posted on 3/29/2026.

Summary

SFCL Tech is hiring a results-oriented Senior Machine Learning Engineer to own anomaly-detection solutions for variables in Battery Energy Storage Systems (BESS). You will design, build, validate, and deploy machine-learning models and pipelines that continuously monitor BESS telemetry (cell voltages, currents, temperatures, state-of-charge, inverter metrics, etc.), detect deviations from normal behavior, support root-cause analysis, and drive operational alerts and remediation. An electrical engineering background is a strong plus.

Key Responsibilities

• Lead the end-to-end development of ML solutions for anomaly detection in BESS telemetry: problem framing, feature engineering, modeling, validation, deployment, and lifecycle maintenance.

• Design and implement unsupervised / semi-supervised / supervised anomaly-detection methods appropriate for time-series and multivariate signals (autoencoders, variational autoencoders, sequence models, isolation forest, one-class models, probabilistic models, change-point detection, etc.).

• Build robust data pipelines for streaming and batch telemetry (Kafka, Spark, Dataflow, or equivalent), including preprocessing, labeling strategies for sparse labels, and synthetic/anomaly augmentation where needed.

• Work with BMS, controls, and asset teams to translate domain knowledge (SOC, SOH, impedance, thermal dynamics, inverter behavior) into features, baseline models, and actionable alert rules.

• Validate models with realistic testbeds and field data, define detection thresholds, and quantify detection performance with business-relevant metrics (precision@k, recall, time-to-detect, false alarm rate, cost of missed events).

• Integrate models into production monitoring systems and dashboards; implement model serving, retraining, and monitoring (ML observability).

• Conduct root-cause analysis for detected anomalies and produce clear findings and remediation guidance for operations and engineering stakeholders.

• Create clear documentation and model governance: assumptions, failure modes, performance over time, and retraining criteria.

• Mentor junior engineers and collaborate cross-functionally with data engineering, controls, operations, safety, and asset management teams.

Required Qualifications

• Bachelor’s or Master’s degree in Computer Science, Data Science, Electrical Engineering, or related STEM field.

• 5+ years of applied machine-learning experience, with at least 3+ years working on anomaly detection or time-series ML in production.

• Strong experience with time-series and multivariate anomaly detection techniques (autoencoders/LSTMs/Transformers for forecasting/reconstruction; probabilistic methods; isolation forest; change-point detection).

• Proficient in Python and ML libraries (PyTorch or TensorFlow/Keras, scikit-learn, pandas, numpy). Experience with model serving frameworks (TF Serving, TorchServe, BentoML, Seldon) is highly desirable.

• Hands-on experience building data pipelines (streaming and batch) and working with telemetry, ideally using Kafka, Spark, Flink, or cloud streaming services.

• Demonstrated experience deploying ML models to production and implementing model monitoring, retraining pipelines, and CI/CD for ML (MLOps).

• Strong quantitative skills: statistical modeling, uncertainty estimation, evaluation metrics for rare events, and A/B or backtest methodology for anomaly systems.

• Excellent communication skills and proven ability to translate technical results into operational actions for multidisciplinary teams.

Preferred / Nice-to-Have

• Bachelor’s or Master’s in Electrical Engineering, or equivalent experience in power systems, power electronics, or controls. (This is a plus—helps bridge model outputs to electrochemical and inverter physics.)

• Experience specifically with Battery Energy Storage Systems (BESS): battery management systems (BMS), SOC/SOH estimation, thermal modeling, cell balancing, inverter/PCS signals, CAN/Modbus/SCADA integrations.

• Experience with physics-informed ML, digital twins, or hybrid model approaches (combining first-principles and data-driven models).

• Familiarity with domain standards and protocols used in energy systems (IEC, IEEE, SCADA protocols) and safety/compliance considerations.

• Cloud experience (AWS/GCP/Azure), including serverless and containerized deployments.

• Experience with visualization tools and dashboards (Grafana, Kibana, Power BI) and building alerting rules integrated into operations tools.

• Experience working with imbalanced labels and developing evaluation strategies for rare-event detection.

• Prior experience in an energy, utilities, renewables, or industrial IoT environment.

Ideal Candidate Profile (short)

• A pragmatic ML engineer who combines deep hands-on expertise in anomaly detection for time-series with strong production deployment experience, and who can rapidly translate BESS domain knowledge into robust detection systems. Ideally has an electrical engineering foundation or close collaboration experience with BESS/control engineers and is comfortable communicating findings to operations, safety, and management teams.

Interview / Screening Guide

Screening questions

1. Describe a production anomaly-detection project you led. What was the signal, which algorithms did you try, and how did you evaluate success?

2. Which approaches do you prefer for anomaly detection when labels are scarce? Explain pros/cons and a real example.

3. Have you worked with BESS or other power electronics telemetry? Which variables were most useful for detection and why?

Technical questions / take-home ideas

• Design an anomaly detection pipeline for multivariate BESS telemetry (cell voltages, currents, temperatures, inverter current/power). Include preprocessing, model choice, thresholding strategy, and deployment considerations.

• Coding exercise: Given a multivariate time series snippet, implement a simple LSTM-autoencoder in Python that outputs an anomaly score per time step. Provide a short evaluation on synthetic anomalies and show how you’d choose a detection threshold.

• System design: Architect a scalable real-time detection system that ingests telemetry from hundreds of BESS sites, performs anomaly scoring with <10s latency, and routes alerts into the operations dashboard. Discuss tradeoffs (latency vs. complexity, centralized vs. edge detection).

Evaluation criteria

• Depth of ML and time-series expertise.

• Practical experience shipping and operating models (MLOps).

• Domain fluency in power or electrical systems (or the ability to learn quickly).

• Clear, actionable thinking about false positives/negatives and business impact.

• Communication skills and cross-team collaboration.

KPIs / Success Metrics (first 6–12 months)

• Reduction in undetected critical BESS events (quantified against baseline).

• False-alarm rate lowered to an agreed-upon operational threshold while maintaining detection recall.

• Successful deployment of at least one real-time anomaly detection model integrated into operations dashboards and alerting.

• Documented procedures and retraining workflows for model lifecycle management.

• Evidence of root-cause identification leading to operational improvements or safety enhancements.