Data Scientist

We are working with years of SCADA telemetry and industrial sensor data covering traction power, tunnel ventilation, signalling and interlocking events, rolling stock condition monitoring, environmental systems, and building systems. The goal is to move from dashboarding and post-incident analysis to predictive operations, forecasting demand, detecting anomalies, and predicting equipment failures before they happen.

Alongside time-series modelling, this role will build applied GenAI solutions that make operational and maintenance knowledge easier to access across incident reports, work orders, O& M manuals, engineering standards, and other unstructured sources. The end users are engineers and control-room teams, so model reliability, explainability, and practical usability are as important as raw accuracy.

Responsibilities:

• Build forecasting models for energy demand, traction power, load profiles, and operational planning.
• Develop anomaly detection models for multivariate sensor streams across point machines, HVAC, traction equipment, track circuits, and other assets.
• Build predictive maintenance and Remaining Useful Life (RUL) models using condition-monitoring data.
• Work with reliability and domain engineers to connect models with real-world failure modes.
• Handle industrial data challenges, including irregular sampling, deadband compression, sensor drift, missing or stale tags, label scarcity, and highly rare failures.
• Apply signal processing where appropriate, including spectral analysis, wavelets, and feature extraction from vibration and waveform data.
• Build RAG systems over maintenance records, incident reports, O& M manuals, and engineering standards.
• Use LLMs for structured extraction from work orders, fault logs, shift reports, and other operational text.
• Convert unstructured operational information into usable labels and features.
• Prototype natural-language access to operational data, including text-to-SQL and semantic query layers over the data lake.
• Build systematic evaluation frameworks covering accuracy, grounding, refusal behavior, and reliability.
• Define clear boundaries for advisory-only GenAI applications in operational and OT environments.
• Partner with data engineers on feature pipelines and with data scientists and OT engineers on domain validation.
• Deploy and monitor models in production.
• Track model drift, degradation, and retraining triggers.
• Communicate model results, assumptions, and uncertainty to non-technical stakeholders.
• Document modeling decisions, limitations, and assumptions.

Requirements:

• 4+ years of applied data science experience, with models deployed in production.
• Strong time-series experience with real-world sensor and telemetry data.
• Hands-on knowledge of classical time-series methods such as ARIMA, SARIMAX, ETS, and state space models.
• Experience with gradient boosting using engineered time-series features.
• Experience with deep-learning approaches such as LSTM, GRU, TCN, or temporal transformers.
• Strong experience in anomaly detection and forecasting on sensor and telemetry data.
• Hands-on GenAI and LLM application development experience.
• Strong knowledge of RAG, embeddings, vector databases or vector search, prompt engineering, and structured outputs.
• Experience building systematic LLM and GenAI evaluation frameworks.
• Expert-level Python, pandas, NumPy, scikit-learn, and PyTorch or TensorFlow.
• Strong SQL skills.
• Strong statistical fundamentals, including temporal validation, data leakage prevention, uncertainty quantification, and model evaluation.
• Experience with MLOps, including experiment tracking, model and version management, and production monitoring.
• Experience with MLflow, Weights and Biases, or similar tools.
• Ability to work directly with domain and engineering teams and translate domain knowledge into modelling features and constraints.

Good To Have:

• Industrial, OT, or IoT sensor data experience.
• Experience with SCADA, industrial historians, and tag hierarchies.
• Railway or metro domain experience, especially rolling stock condition monitoring, traction power, signalling assets, delay or headway analysis, and energy optimization.
• Predictive maintenance or reliability engineering experience.
• Knowledge of RCM, FMEA, RAMS, or survival analysis.
• Experience with time-series foundation models such as Chronos, TimesFM, or Moirai.
• Experience with LLM fine-tuning, agent frameworks, guardrails, or AI safety tooling.
• Experience with model explainability and deploying models for operational users.
• Experience working in regulated engineering environments.
• Experience with real-world sensor and telemetry data rather than only structured business datasets.
• Comfort taking models from data exploration through modelling, evaluation, production deployment, and monitoring.