Hemlata Kumari

<p>Supporting Model Risk Management (MRM) implementation, strengthening model governance practices and ensuring model documentation aligns with regulatory standards following SS1/23 model risk governance guidelines.<br>
Contributing to model governance frameworks across the model lifecycle, ensuring transparency, auditability and regulatory compliance.<br>
Conducting model documentation review, methodology validation and governance checks.<br>
Led and mentored junior analysts and modelers in developing end-to-end credit risk scorecards using both machine learning and traditional statistical techniques.<br>
Provided guidance across the complete model lifecycle, including data preparation, feature engineering, model development, validation, monitoring and documentation.</p>

<p>Designed and implemented Machine Learning–based credit scorecards, including development of a Waython-based ML behavioral scorecard for the Singapore PLOC project and Malaysia Credit Card portfolio, covering project design, data preparation, feature engineering, model development and validation.<br>
Collaborated closely with data team to define new data structures for the first PLOC product-based modeling initiative, addressing data definition gaps and ensuring accurate feature engineering inputs for model development.<br>
Performed strategic insights using vintage analysis, roll rate analysis and loss performance tracking.<br>
Performed model performance validation using key metrics including KS, Gini, ROC, PSI (population stability) and model accuracy checks, ensuring model robustness and predictive performance across segments.<br>
Collaborated with the risk strategy team to assess the impact of transitioning from a decision tree–based cut-off framework to a machine learning–based model, evaluating changes in portfolio loss metrics.<br>
Maintained the same score cut-off used in the decision tree approach to perform a controlled comparison and quantify the incremental benefit of the machine learning model.<br>
Partnered with implementation teams to support model deployment, UAT testing and credit decision engine implementation.<br>
Worked with technology teams to understand decision engine architecture and deployment workflows, supporting model implementation planning.<br>
Built portfolio monitoring and model performance frameworks to track model drift, stability and portfolio risk trends, enabling early detection of deteriorating credit performance.<br>
Led bureau performance analytics and validation, including the CBH-CRIB Bureau Validation project, performing detailed performance assessment and aligning bureau model performance with internal credit risk modeling standards.<br>
Evaluated the incremental value of incorporating credit bureau scores into an existing behavioral risk model, conducting cost-benefit analysis given the cost sensitivity of bureau data.<br>
Led end to end in developing HSBC’s internal Python-based automated strategy decision tool, performing comparative testing against manual strategy execution and achieving 2× improvement in processing efficiency with equivalent results.</p>

<p>Built credit risk scorecards and predictive models across the customer lifecycle (origination, behavior, collections and recovery) for retail portfolios in the EMEA region, applying WOE binning, segmentation, logistic regression and reject-inferencing techniques to improve risk predictiveness and underwriting decisions.<br>
Built predictive models using SAS & Python, performing data sourcing, cleansing, exclusion & validation, feature engineering, model training, validation and comprehensive documentation.<br>
Applied traditional & advanced machine learning algorithms including Generalized Additive Models (GAM), Decision Trees and Random Forest to improve predictive performance while balancing model interpretability.<br>
Supported model deployment, validating production datasets and score distributions to ensure accurate model execution within decision systems.<br>
Conducted deployed model performance evaluation and monitoring, generating score distribution analysis and model monitoring reports to track the predictiveness and stability.<br>
Performed scorecard re-weighting & recalibration, ensuring models remained aligned with evolving portfolio risk profiles and changing macro-economic conditions.<br>
Partnered with banking clients and internal analytics teams to gather requirements and translate business risk needs into scalable predictive modelling solutions; aligned timeline & expectations, clarified data gaps, discussed progress along with modelling choices at each stage, presented findings and provided guidance on industry best practices.<br>
Key contribution to the development of reusable notebook templates on the FICO Analytic Workbench platform for scorecard development and machine learning model deployment, enabling scalable and standardized credit risk analytics workflows across teams.<br>
Reviewed and optimized notebooks for user experience (ease of use, UI, explainability) and analyst requirements (output accuracy, model interpretability and business purpose alignment) while conducting external benchmarking to enhance model development concepts and resource management efficiency.</p>

<p>Analyzed raw test data using distribution techniques and statistical modeling to identify material specifications and derive allowable material limits.<br>
Performed t-tests and correlation analysis to identify critical materials and suppliers; built predictive models for key components using linear regression.</p>

<p>Led a project to evaluate and implement a new pot process control framework, managing analysis, experimentation, and deployment.<br>
Assessed live factory control systems, conducted technical feasibility studies, and executed pot-level experiments to generate and analyze test data.<br>
Applied advanced statistical analysis to develop real-world predictive insights and implemented a new control logic based on test behavior.<br>
Performed iterative testing using linear regression and time-series analysis on factory data to optimize the control logic and improve process efficiency and operational metrics.</p>