Job Description

AI ML Consultant

Sr. Data Scientist — Manufacturing & Process AI

Location: Delhi NCR

About the role

You will build and deploy machine-learning models directly on plant data to cut energy

consumption, improve equipment reliability, and tighten product quality across our cement

operations. This is a hands-on modelling role embedded with process, operations and reliability

teams — your work will be measured in real, finance-validated savings (kcal/kg clinker,

kWh/tonne, avoided downtime), not slide decks.

Key responsibilities

• Build, validate and deploy ML models for process optimisation (kiln / pyro-process control,

grinding & separator efficiency), predictive maintenance on critical rotating equipment,

and quality / clinker-factor optimisation.

• Work with high-frequency sensor and time-series data from plant historians, DCS and IIoT

systems; engineer meaningful features from noisy, real-world industrial signals.

• Partner with plant operators and process engineers to encode domain knowledge into

models, and to take models safely from advisory recommendations toward closed-loop

control.

• Establish rigorous baselines and quantify impact with finance-grade discipline; defend

results under scrutiny.

• Work with the MLOps / platform team to productionise models and monitor them in live

operation.

• Communicate findings clearly to non-technical plant leadership.

Required qualifications (must-have)

• Bachelor's or Master's in Engineering (Chemical, Mechanical, Electrical, Industrial),

Statistics, Computer Science, or a related quantitative field.

• 3–6 years building and deploying ML models, including demonstrable experience in a

manufacturing or process-industry environment (cement, steel, refining, chemicals,

power, glass, mining, or similar).

• Strong applied skills in time-series analysis, sensor/signal data, anomaly detection,

regression and forecasting, with a solid statistics foundation.

• Strong, idiomatic Python for data science (NumPy, pandas, SciPy, scikit-learn,

statsmodels) with clean, tested, production-quality code; strong SQL.

• Deep command of classical / traditional machine learning — regularised regression

(Ridge, Lasso, ElasticNet), tree-based ensembles (Random Forest, Gradient Boosting —

XGBoost / LightGBM / CatBoost), SVM, k-NN and Naive Bayes — with sound feature

engineering, cross-validation and hyperparameter tuning.

• Proven ability to wrangle messy industrial data and engineer features that work in

production.

• Comfortable on the plant floor — explaining models to engineers and operators and earning

their trust.

Preferred (strong pluses)

• Hands-on experience with Industrial IoT (IIoT) and Operational Technology (OT) data —

plant historians (OSIsoft PI / AVEVA, Aspen IP.21), OPC-UA, SCADA / DCS, time-series

databases.

• Domain exposure to cement or heavy/process manufacturing (pyroprocessing, grinding,

combustion, quality control).

• Experience working with data from SAP (ERP — especially PM / PP / production &

maintenance modules) and Salesforce (SFDC).

• Familiarity with Advanced Process Control (APC) concepts and closed-loop deployment.

• Deep learning for time series; physics-informed or hybrid (data + first-principles) modelling.

Technical skills

• Programming & engineering: idiomatic, production-quality Python — NumPy, pandas and

SciPy for vectorised data work; clean, modular code with unit tests (pytest); OOP; virtual

environments & packaging; Jupyter; Git. Strong SQL; PySpark for large datasets a plus.

• Classical machine learning: hands-on depth across regularised regression, tree-based

ensembles (Random Forest, XGBoost / LightGBM / CatBoost), SVM, k-NN and Naive

Bayes; unsupervised methods — k-means, DBSCAN, hierarchical clustering and PCA /

dimensionality reduction.

• Statistical & modelling rigour: hypothesis testing, regression diagnostics, feature

engineering & selection, cross-validation, hyperparameter tuning, class-imbalance handling,

and disciplined error analysis.

• Time-series & anomaly detection: classical methods (ARIMA / SARIMA, exponential

smoothing, state-space models) and libraries (statsmodels, sktime, tsfresh, Prophet);

anomaly detection (Isolation Forest, One-Class SVM).

• Core libraries: scikit-learn, statsmodels, XGBoost / LightGBM, matplotlib / seaborn.

Platform & tooling

Cloud / lakehouse (Azure, AWS or Databricks); plant historian & OT connectors; Git-based

workflows.

r