Machine Learning Engineer/Data Scientist (Forecasting)

Position Summary

We are seeking a highly skilled Machine Learning Engineer / Data Scientist specializing in Forecasting to design, develop, deploy, and continuously improve machine learning solutions supporting an enterprise AI/ML analytics platform.

The role will focus heavily on time-series forecasting, predictive modeling, feature engineering, model validation, MLOps, and production deployment. The successful candidate will work directly with large-scale operational and healthcare datasets to build forward-looking models for patient volume, capacity utilization, revenue and collections, and other operational metrics.

This is a hands-on model-building and productionization role, requiring ownership of the complete ML lifecycle, from raw data and feature engineering through model development, backtesting, deployment, monitoring, and automated retraining.

The role will also contribute to leadership dashboards and AI-assisted analytics by ensuring forecasts, scenarios, and narrative insights are grounded in validated model outputs and governed business metrics.

Key ResponsibilitiesForecasting & Predictive Modeling • Design and develop production-grade time-series forecasting models for patient and appointment volume, capacity utilization, revenue/RCM performance, and other operational domains. • Develop forecasts across daily and weekly horizons using appropriate statistical and machine learning techniques. • Build and compare statistical forecasting approaches including: • Seasonal naive and other baseline models • Exponential smoothing / ETS • ARIMA, SARIMA and SARIMAX • STL decomposition • Prophet or equivalent additive forecasting models • Develop machine-learning-based forecasting models using LightGBM, XGBoost or similar gradient-boosting techniques with lag, calendar, capacity and operational features. • Incorporate multiple seasonalities, holidays, clinic closures, provider schedules, PTO and other known-future variables into forecasting models. • Produce probabilistic forecasts and prediction intervals, providing leadership with realistic ranges rather than point estimates alone. • Address cold-start scenarios, structural changes, new providers, new clinics, sparse series and regime changes. • Support hierarchical forecasting requirements across clinic, department, provider and visit-type levels where required. • Develop scenario and what-if forecasting capabilities to support operational and leadership decision-making.

Data & Feature Engineering • Profile and validate source datasets before model development, including data coverage, duplicates, missing records, inconsistent keys and competing data sources. • Build reusable and reproducible feature engineering pipelines from cloud data platforms into modeling-ready datasets. • Develop features including: • Calendar and holiday variables • Lag features • Rolling averages and rolling windows • Booking-curve and appointment-fill features • Capacity and available-slot indicators • Provider and department hierarchies • Telephony leading indicators • Patient-panel and care-gap measures • Ensure point-in-time correctness and prevent future information from leaking into model training. • Identify and resolve real-world data issues including inconsistent provider or department naming, missing days, restated historical records, cancellation text, backfilled data and partial loads. • Optimize SQL and data pipelines through partition filtering, efficient column selection, intermediate tables and appropriate data-modeling practices.

Model Validation & Performance Management • Develop rigorous rolling-origin / expanding-window backtesting frameworks appropriate for time-series data. • Evaluate model performance across multiple forecast horizons rather than relying on a single aggregate accuracy score. • Benchmark every production model against appropriate baseline methods including: • Seasonal naive • Trailing averages • Prior-year comparable periods • Use appropriate forecasting metrics including MAE, RMSE, WAPE/WMAPE, MASE, bias/mean error and pinball loss. • Ensure models demonstrate measurable improvement over baseline performance before being promoted to production. • Identify and eliminate data leakage through feature-availability and as-of-date validation. • Track forecast-versus-actual performance by business domain and forecast horizon. • Use model explainability techniques such as SHAP and feature importance to identify key forecast drivers and translate findings into business language.

Predictive / Classification Modeling • Support development of predictive models including: • Appointment no-show risk • Cancellation risk • Denial likelihood • Payment-delay risk • Develop calibrated probability models appropriate for operational decision-making. • Evaluate classification models using precision, recall, PR-AUC and probability-calibration techniques. • Define operational thresholds based on the business impact of false positives and false negatives rather than relying solely on technical model metrics.

MLOps & Productionization • Own the end-to-end lifecycle of production ML solutions, including training, validation, deployment, monitoring and retraining. • Establish experiment tracking and model-registry processes using MLflow, Weights & Biases or equivalent technologies. • Build automated pipelines covering:Feature Preparation → Model Training → Backtesting → Accuracy Validation → Model Registry → Production Promotion → Forecast Publishing • Implement automated retraining triggered by schedules, new data availability or model-performance deterioration. • Establish accuracy gates to ensure that underperforming models are not automatically promoted to production. • Version datasets, features and models to ensure every published forecast is fully reproducible and auditable. • Monitor: • Forecast accuracy • Model drift • Feature and data drift • Data freshness • Pipeline failures • Build model-health monitoring and alerting processes. • Develop live model-serving APIs using FastAPI, Flask or equivalent frameworks for forecasts and scenario analysis. • Develop batch scoring processes to publish daily forecasts to downstream applications. • Support containerization and automated deployment using Docker and CI/CD pipelines.

AI / LLM Integration • Integrate forecasting capabilities with AI assistants and analytics applications. • Work with GPT, Claude or similar LLM APIs, including structured outputs, schema validation, system prompting and function/tool calling. • Ensure AI-generated narrative explanations are grounded in actual model outputs and governed metrics. • Prevent AI systems from generating unsupported or fabricated numerical results. • Support development of governed semantic layers and read-only analytical interfaces. • Contribute to tool-routing, multi-step AI workflows and contextual AI-assistant capabilities. • Support automated narrative generation for forecast variances and leadership reporting where required.

Visualization & Executive Analytics • Publish model outputs in formats consumable by dashboards, semantic layers and downstream analytics applications. • Develop or support leadership dashboards using Power BI or Tableau. • Build Python-based internal analytics applications using Streamlit, Dash, Flask or equivalent tools. • Create forecast-versus-actual visualizations using Plotly, Matplotlib or similar libraries. • Present forecast uncertainty through prediction ranges, fan charts and clearly identified actual, booked and predicted values. • Translate technical model findings into concise, actionable recommendations for operational and executive stakeholders. • Support forecast-accuracy and variance reviews with senior management.

Required Qualifications & Experience • 3–5 years of professional experience in Data Science, Machine Learning, ML Engineering or analytics-focused software engineering. • Candidates with approximately 2 years of strong hands-on experience and demonstrable production forecasting expertise may also be considered. • Proven experience personally building, deploying and maintaining at least one end-to-end production machine learning application. • Hands-on professional experience in time-series forecasting is mandatory. • Strong experience working with large, complex and imperfect real-world datasets. • Demonstrated ability to identify whether poor model performance originates from data quality, feature engineering or model design. • Strong ability to translate business questions into analytical or machine learning solutions. • Ability to explain technical model outputs in clear operational and executive language. • Ability to independently own a technical domain in a fast-moving, high-accountability environment. • Ability to become productive quickly with limited ramp-up time.

Required Technical Skills: Programming & Data • Advanced Python, including: • pandas • NumPy • Modular and testable Python development • Vectorized data processing • Dependency and environment management • Git-based development workflows • Advanced SQL, including: • Window functions • CTEs • Complex joins • LAG / LEAD • ROW_NUMBER • Running aggregates • As-of/date-range joins • Multi-grain aggregation • Query optimization and partition awareness • Strong understanding of: • Data modeling • Data grain • Layered / medallion architectures • Idempotent data processing • Slowly changing datasets

Forecasting

Strong hands-on knowledge of several of the following: • ARIMA / SARIMA / SARIMAX • ETS / Exponential Smoothing • STL decomposition • Prophet • LightGBM • XGBoost • Lag-based forecasting • Global / pooled forecasting models • Multi-step forecasting • Probabilistic forecasting • Quantile forecasting • Prediction intervals • Multiple seasonality • Known-future covariates • Cold-start and structural-change handling

Familiarity with at least two forecasting libraries such as: • statsmodels • Prophet • sktime • Nixtla StatsForecast / MLForecast / NeuralForecast • Darts • PyTorch Forecasting • pmdarima

MLOps

Hands-on experience with: • MLflow, Weights & Biases or equivalent model-management platforms • Automated retraining • Model registries • Model monitoring • Drift detection • Model reproducibility • FastAPI or Flask • Batch scoring • Model-serving APIs

Exposure to the following is strongly preferred: • Docker • GitHub Actions • Azure DevOps or similar CI/CD solutions • Optuna, Hyperopt or similar hyperparameter-optimization frameworks • Airflow, Prefect, Dagster, ADF or comparable orchestration platforms

Cloud & Analytics Platforms

Experience or exposure to: • AWS S3 • Athena • AWS Glue • Microsoft Fabric • Synapse • SQL data warehouses

Experience with Athena / Presto / Trino SQL dialects is preferred.

AI / LLM

Hands-on knowledge of: • GPT, Claude or comparable LLM APIs • Structured / JSON outputs • Schema validation • Function and tool calling • Grounded AI generation • Prompt design • Token and API-cost management

Understanding of text-to-SQL, semantic layers, agent workflows and LLM evaluation is highly desirable.

Business Intelligence & Visualization • Power BI or Tableau • Streamlit, Dash or Flask • Plotly and/or Matplotlib • Executive-facing analytical and forecasting dashboards

Healthcare & Business Domain

Previous healthcare experience is preferred but not mandatory.

The successful candidate will be expected to quickly develop familiarity with: • Healthcare practice and appointment operations • Provider scheduling and capacity management • Patient no-shows and cancellations • Revenue Cycle Management • Claims, denials and AR • Value-Based Care • HCC / RAF concepts • Healthcare quality measures and care gaps • RPM / CCM • Transitional Care Management

The candidate must demonstrate disciplined handling of sensitive healthcare information, including appropriate use of PHI, minimum-necessary access, de-identification, audit logging and role-based access controls. HIPAA training will be provided as required.

Preferred / Good-to-Have Experience • Previous experience with healthcare, payer, claims, insurance or financial time-series data. • Experience with dbt, Snowflake, Databricks or Spark. • Optimization, workforce scheduling, queueing or discrete-event simulation experience. • Causal inference or uplift-modeling experience. • Experience measuring the impact of operational interventions. • Familiarity with Smartsheet or similar workflow and reporting platforms. • Experience working directly with senior leadership or C-level stakeholders. • Familiarity with deep-learning forecasting architectures such as N-BEATS, NHITS, Temporal Fusion Transformer or DeepAR, together with the judgment to determine when simpler forecasting approaches are more appropriate.

Key Success Measures

Success in this role will be demonstrated through: • Reliable forward-looking forecasts with clearly communicated confidence ranges. • Forecast models that consistently outperform documented baseline models. • Production pipelines that execute automatically and reliably. • Transparent forecast-versus-actual accuracy reporting. • Effective detection and alerting of data, model and pipeline issues. • Fully reproducible and auditable model outputs. • Forecasts that can be directly consumed by leadership dashboards and AI-assisted analytics. • Clear documentation of model assumptions, limitations, deployment processes and operating procedures.

Pay: Rs1.00 - Rs2.00 per month

Work Location: In person