Mounika T

Data Scientist

Bengaluru 3+ yrs exp 87 · Excellent

About

Results-driven Data Scientist with 3.5 years of experience building end-to-end ML models across credit risk, collections, revenue analytics, and customer intelligence. Proven track record of delivering measurable business outcomes — including 5% improvement in approval ratios, 3% reduction in predicted default rates, and ~50% cost savings on customer interventions. Combines deep technical expertise in Python, SQL, ML, feature engineering, model validation and explainability.

Skills & Expertise (20)

Python Advanced

8.5/10

3.5

Years Exp

SQL Advanced

8.0/10

3.5

Years Exp

Model validation Advanced

7.6/10

3.5

Years Exp

Logistic Regression Advanced

7.5/10

3.5

Years Exp

Random Forest Advanced

7.5/10

3.5

Years Exp

XGBoost Advanced

7.5/10

3.5

Years Exp

SVM SHAP scikit-learn Matplotlib Pandas NumPy KNN Naive Bayes Vintage Analysis REST APIs Deployment GridSearchCV ROC-AUC Roll Rate Analysis

Work Experience

Data Scientist

Soft Suave Technologies

Jan 2023 - Jul 2026

Built Application Scorecard model from scratch: Roll Rate Analysis to define the default event; Vintage Analysis to set the performance window; NTC (New-to-Credit) sub-model for seamless onboarding of thin-file customers. Improved approval ratio by 5% while simultaneously reducing predicted default rate by 3%. Extracted, joined, and transformed customer and application-level data using SQL; built a rich feature universe spanning bureau data, banking variables, and alternative data sources; applied WOE/IV, CSI, and VIF for disciplined feature selection. Trained and tuned Logistic Regression, Random Forest, and XGBoost models using GridSearchCV and Optuna; evaluated model discrimination and stability with KS, Gini, ROC-AUC, and PSI. Implemented Reject Inferencing to maintain approval ratio integrity and SHAP explainability for regulatory and business transparency. Built supporting assets: Income Estimation model (FOIR-based Amount Strategy), Ready Reckoner for instant and upfront approve/reject decisions, and cloud-deployed REST APIs (Python/JSON). Designed a suite of 3 targeted collection scorecards, enabling precise intervention at each delinquency stage. Built Behavioural Scorecard for the Personal Loans portfolio, serving dual purposes: ECL (Expected Credit Loss) computation (PD/LGD/EAD) and identification of upsell/top-up candidates. Achieved ~50% reduction in customer visit and intervention costs. Architected a propensity model to identify high-probability cross-sell opportunities, improving targeting precision and reducing outreach costs. Developed a forward-looking LTV model using historical transaction and behavioral data for targeted marketing and increased profitability. Built a multi-feature classification model to identify at-risk customers, enabling proactive retention campaigns and measurably reducing attrition. Developed a corporate PD (Probability of Default) model at origination using financial statement data, credit bureau variables, and application attributes for risk-based underwriting decisions. Built a lead scoring model to rank prospects by conversion likelihood, improving sales efficiency and measurably increasing conversion rates.