Data Scientist/ML

Job Summary
Photon is hiring an experienced Data Scientist / Machine Learning Engineer, to design, build, validate, and operationalize a closed-loop Propensity-to-Pay model for Enercare Collections. The engagement runs on Enercares existing Azure platform (Azure Data Factory, ADLS, Azure Machine Learning, and Azure SQL) and reuses current ingestion and CDS integration patterns.

The role owns one primary Collections model: define the pay target and scoring population, engineer features from customer, account, payment, AR, invoice, usage, delinquency, write-off, and collections history, rank accounts by likelihood to pay, and assign agreed risk segments. Scores are produced on a scheduled batch cycle, with model version and evaluation details published for CDS. Outcome and treatment feedback are brought back so cycle-over-cycle performance can show when the model should be retrained.

This is a hands-on delivery role. The hire is expected to contribute from Discovery and to take the model through validation, UAT, production scoring, and handover with in agreed timeline with customer.

Key Responsibilities

- Work with Collections business owners and data owners during Discovery to confirm the scoring population, the definition of “pay,” the prediction horizon, label history, success metrics, and acceptance thresholds.
- Assess whether available history, quality, and coverage can support a viable model, and document limitations, assumptions, and the baseline comparison against the current Collections prioritisation process.
- Define and build payment, AR, billing, customer, usage, delinquency, write-off, and collections behavioural features from the consolidated analytical dataset. Freeze the feature list with the data science lead and business owner before model build.
- Build and train one primary Propensity-to-Pay model, including target construction, feature selection, training, scoring, and tuning against agreed metrics.
- Validate technically and with the business: scoring completeness, data-leakage controls, reproducibility, output consistency, ranking and segment behaviour, calibration, and lift versus the agreed Collections baseline.
- Convert scores into risk segments using agreed thresholds, retain the model segment when an authorized business override is applied, and support treatment/action mapping defined by Collections.




- Produce a model card, validation evidence, and a reviewable scored sample for stakeholder sign-off.
- Implement model governance on Azure Machine Learning: unique model version identifiers, evaluation results, visibility of which version is deployed in each setting, and which version produced each scoring output.
- Support an alternate-version path so a candidate model can be tested in the operations layer before approval, and support controlled rollback to the previous approved version.
- Operationalize scheduled batch scoring with feature refresh, score generation, and controls that publish only complete, validated score sets. Failed runs must leave the last valid published scores in place.
- Design the closed loop: capture operational feedback and actual pay outcomes, compare performance across cycles, and recommend when retraining is required. Document what would be needed for later automated retraining; implementation beyond the agreed Collections scope is a separate change.
- Collaborate with data engineers on ADF, ADLS, and Azure SQL pipelines, and with QA on reconciliation of counts, score distribution, segments, and CDS output fields (customer/account key, propensity score, segment, model version, evaluation date).
- Use Enercare’s Azure DevOps promotion path across Development/QA, UAT, and Production. Support UAT, defect resolution, hypercare, runbook handover, and knowledge transfer.
- Document features, experiments, metrics, thresholds, overrides, and operating guidance so Enercare operations can run a production scoring cycle with guidance.

Required Skills and Qualifications

- Bachelor’s or Master’s degree in Computer Science, Data Science, Statistics, Mathematics, or a related field.
- 5–8 years in data science or machine learning, including at least one classification or ranking model taken through batch scoring in production. This is a lateral hire; campus or training-only profiles will not fit the timeline.
- Strong Python,



with day-to-day use of pandas, NumPy, scikit-learn, and experiment tracking. Ability to read and write SQL against large relational datasets.
Solid statistics and model evaluation for imbalanced collections-style problems: target leakage, train/validation design, precision and recall at operational cut-offs, lift, calibration, stability, and segment behaviour.
- Hands-on Azure Machine Learning for training, model registration, versioning, and batch scoring, with data read from ADLS or Azure SQL.
- Experience shaping features from payment, billing, receivables, or delinquency history, and explaining ranked outputs to business stakeholders who own treatment rules.
- Experience writing the evidence a UAT sign-off needs: metric results, baseline comparison, scored samples, and known limitations.
- Clear written and spoken communication with business owners, data engineers, QA, and operations. Able to work in a fixed-scope, milestone-based delivery with Discovery, UAT, and production gates.

Preferred Skills

- A prior propensity, credit-risk, collections, or churn model where scores drove a treatment or contact strategy.
- Closed-loop model monitoring: outcome labels returning after a decision, drift or performance review, and a documented retrain trigger.
- Azure Data Factory familiarity sufficient to trace feature-refresh and scoring pipeline dependencies, even if pipeline build sits with data engineering.
- MLflow or Azure ML run history, model cards, and promotion of a registered model through DevOps.
- Power BI or a similar tool for score-distribution and segment review with Collections stakeholders.
- Experience with customer/account key consolidation across more than one source before modelling.

Education

Bachelor’s or Master’s degree in Computer Science, Data Science, Artificial Intelligence, Statistics, Mathematics, or a related discipline.

What We Offer

- A defined Collections engagement with a production batch-scoring outcome, CDS consumption, and a closed-loop retraining recommendation, delivered on Azure.
- Direct work with data engineering, QA, DevOps, and the client Collections and CDS owners through Discovery, UAT, and hypercare.
- A lateral role with ownership of the model, the validation evidence, and the handover, inside a fixed 14-week program.

📌 Data Scientist (Chennai)
🏢 Photon
📍 Chennai