Paramount HR
Senior Data Scientist - Credit Risk Modelling
Senior data scientist in Mumbai who owns end-to-end development of credit scorecards and decision models for a lender — defining populations and targets, engineering features, validating champion-challenger models (KS, Gini, PSI) and pushing them into credit policy and decision engines. Core stack is Python, SQL, gradient boosting (XGBoost/CatBoost/LightGBM), MLflow and MLOps.
- Compensation
- Not disclosed
- Location
- Mumbai
- Work setup
- Not specified
- Seniority
- Senior
- Employment
- Not specified
- Experience
- 5+ years
- Category
- Data science
- Posted
Skills
Must have
- 5-8 years hands-on Data Science / statistical modelling experience, including 3+ years in lending, credit risk, underwriting or related BFSI analytics
- Strong Python and SQL
- Personally built at least one credit scorecard or underwriting model end-to-end (population/target design, feature engineering, validation, calibration, deployment, monitoring)
- Strong understanding of bureau and lending data, model governance, explainability, reason codes and production implementation via policy/BRE, APIs or decision engines
- Exposure to logistic scorecards and ML methods such as XGBoost/CatBoost/LightGBM
- Experience with cloud, MLflow, APIs, Git and MLOps practices
- Understanding of statistical methods: regression, time series, sampling theory, hypothesis testing
- B.E/B.Tech/B.Stat degree
Nice to have
- Experience in MSME, embedded finance, line-of-credit, working-capital, collections, fraud or early-warning analytics
- Practical exposure to NLP/LLM, RAG or document-intelligence use cases with grounding, evaluation and human review
Full description
About the role:
- Own end-to-end development of credit scorecards and decision analytics across bureau, platform, behavioural and portfolio data - from population and target definition through deployment and monitoring.
- Work with Credit, Risk, Underwriting and Technology to convert model outputs into grades, approval treatment, limit, pricing, tenure and reason codes. Applied AI is a selective secondary capability for document, evidence and analytical assistance - not autonomous financial decisioning.
Key Responsibilities:
- Define development populations, observation/performance windows and targets using portfolio maturity, vintage, roll-rate and business context; benchmark existing scores before recommending a new model or recalibration.
- Clean and profile data, engineer interpretable features, prevent leakage and build explainable benchmark and challenger models across bureau, platform, repayment and other approved data.
- Complete champion-challenger selection and validation using KS, Gini/AUC, calibration, stability/PSI, out-of-time and segment performance; create score scaling, grades, reasons, limitations and model documentation.
- Translate selected models into policy/BRE treatment, approval/referral/rejection, limit, pricing and tenure logic; prepare deployment artefacts, golden cases, API/UAT evidence and production-monitoring requirements.
- Develop EWS, collections, fraud/trust, propensity and portfolio analytics, and selectively support grounded NLP/LLM use cases such as document extraction, evidence retrieval and internal risk summaries.
Core Competencies:
- Strong statistical discipline combined with practical credit judgement - able to distinguish predictive lift from leakage, instability or weak business meaning.
- Hands-on ownership mindset: comfortable coding, challenging data, presenting decisions and following models through production monitoring.
- Clear communicator who can explain model behaviour, limitations and business impact to Credit, Underwriting, Technology, management and assurance teams.
- Understanding of detailed data statistics methods like regression, time series, sampling theory, hypothesis testing etc.
Must-Have Requirements:
- 5 - 8 years of hands-on Data Science / statistical modelling experience, including at least 3 years in lending, credit risk, underwriting or closely related BFSI analytics; strong Python and SQL are mandatory.
- Personally built at least one credit scorecard or underwriting model end-to-end, including population/target design, feature engineering, validation, calibration, deployment support and monitoring.