The live app was rapidly prototyped and iterated using the Lovable AI platform. I translated the ML solution into a user‑friendly flow for credit analysts: upload data, review model‑driven risk insights, and generate decisions with clear rationales.
Defined primary user journeys and wireframed screens before styling. Used componentized UI blocks to ensure consistency and fast changes. Validated flows through quick stakeholder demos.
• Score detail view with risk flags, feature contributions, and recommendations
Delivered a usable front‑end for credit review with fast turnaround, aligning UX with the underlying ML capabilities.
• Accelerated delivery from concept to a usable app using Lovable AI.
• Reduced analyst review time through streamlined flows and clear risk rationales.
• Improved adoption via an approachable UI tightly aligned to ML capabilities.
A machine-learning system to predict consumer loan default risk for LoanTap. It trains a logistic regression model with a custom preprocessing pipeline, serves realtime and batch scoring via FastAPI, and provides a Streamlit dashboard for insights, single-record scoring, CSV uploads, and EDA.
• Model training: Logistic Regression with class balancing, cross‑validation, ROC AUC/accuracy reporting, persisted with joblib.
• Preprocessing pipeline: Feature engineering (temporal, risk flags, binning), label encoding, scaling; saved and reused at inference.
• Interactive UI (Streamlit): Executive dashboard, manual and dataset-driven single prediction, CSV upload scoring, processed feature view, and EDA with ROC/PR/calibration/heatmaps.
• Expecting to create risk assessment frameworks to help financial institutions reduce risk while improving customer experience.
• Expecting to develop automated loan approval processes to streamline decision-making.
Utilized supervised learning techniques including Random Forest, Logistic Regression, and Gradient Boosting to build predictive models. Implemented feature engineering to create relevant variables from customer data. Applied cross-validation and hyperparameter tuning to optimize model performance. Used ROC curves and precision-recall metrics to evaluate model effectiveness.
Lower credit losses and a healthier portfolio
Better risk stratification reduces default rates and NPAs
Enables risk-based pricing/limits, improving expected loss and provisions accuracy Faster, scalable lending decisions Automates low-risk approvals, cutting turnaround time and manual effort Increases throughput and conversion while keeping risk in check.