<link>https://porushy.github.io/</link></image><item><title>Algorithmic Lending & Model Governance</h1> <article> <h1>Algorithmic Lending & Model Governance</h1> <p>Fri, 31 Jul 2026 00:00:00 +0000</p> <p>Jun 2026 – Jul 2026</p> <ul> <li>Built credit-scoring models using Logistic Regression and Random Forest on the German Credit (Statlog) dataset.</li> <li>Conducted SHAP-based explainability and fairness/bias analysis to understand how the model predicts.</li> <li>Identified biases shown by the model based on demographic data and mapped the behaviour to Indian digital governance frameworks.</li> </ul> </article> <article> <h1>Credit Risk & Expected Credit Loss (ECL)</h1> <p>Sat, 25 Jul 2026 00:00:00 +0000</p> <p>Jul 2026</p> <ul> <li>Built a Basel-style expected credit loss engine which converts an algorithmic lending model’s output into calibrated PD and combines it with LGD and EAD to produce loan-level ECL.</li> <li>Implemented IFRS-9 three-stage impairment logic and ran LGD sensitivity analysis to test how provisioning shifts under alternative recovery assumptions.</li> <li>Analysed risk-grade deciles and quantified how sharply expected loss concentrates in a small share of the book, which drives a disproportionate share of provisions.</li> </ul> </article> <article> <h1>Financial Market Surveillance</h1> <p>Mon, 20 Jul 2026 00:00:00 +0000</p> <p>Jul 2026</p> <ul> <li>Built an anomaly-detection model based on an unsupervised learning algorithm called Isolation Forest.</li> <li>Detected anomalies in USD-INR exchange-rate data using rolling-volatility features.</li> <li>Detected stress episodes matching real crisis periods (2008 GFC, 2013 taper tantrum, 2020 COVID), along with the recent 2026 rupee depreciation.</li> </ul> </article> <article> <h1>Consumer Complaint Analytics</h1> <p>Fri, 10 Jul 2026 00:00:00 +0000</p> <p>Jun 2026 – Jul 2026</p> <ul> <li>Engineered a full NLP pipeline on the CFPB Consumer Complaint Database (8+ GB) using TF-IDF and logistic regression.</li> <li>Made an automated complaint classifier, which achieved ~0.87 accuracy.</li> <li>Identified the growing share of templated filings and distinguished automated volume from genuine consumer harm using advanced statistical analysis, such as Cramér’s V and the chi-square test of independence.</li> </ul> </article> <article> <h1>Experience</h1> <p>Tue, 24 Oct 2023 00:00:00 +0000</p> </article> <article> <h1>European Option Pricing under Black-Scholes</h1> <p>Mon, 30 Nov 2020 00:00:00 +0000</p> <p>Aug 2020 – Nov 2020 · MS Project</p> <ul> <li>Analysed risk-neutral valuation and delta-hedging of vanilla and binary European options via the Black-Scholes framework, including option Greeks and implied volatility.</li> <li>Simulated a volatility surface for Microsoft stock in R.</li> </ul> </article> <article> <h1>Bayesian Logistic Regression with MCMC</h1> <p>Sat, 30 Nov 2019 00:00:00 +0000</p> <p>Aug 2019 – Nov 2019</p> <ul> <li>Estimated the posterior distribution of a parameter in a given model using Bayesian logistic regression.</li> <li>Used MCMC to implement the Metropolis-Hastings algorithm to draw samples from the posterior.</li> </ul> </article> <article> <h1>Mixture of Gaussians with Cross Validation</h1> <p>Fri, 29 Nov 2019 00:00:00 +0000</p> <p>Aug 2019 – Nov 2019</p> <ul> <li>Predicted the number of hidden classes in given data from a bivariate mixture of Gaussians using cross-validation and the Akaike Information Criterion (AIC).</li> <li>Implemented the EM algorithm to find MLE estimates of a given mixture with the predicted classes in R.</li> </ul> </article> <article> <h1>Optimization Algorithm</h1> <p>Tue, 30 Apr 2019 00:00:00 +0000</p> <p>Jan 2019 – Apr 2019</p> <ul> <li>Developed an algorithm to find the orthogonal projection of a point on a parametric curve, and approximated a given function based on discrete data.</li> <li>Used Newton’s method to find the root, and Chebyshev interpolation by choosing Chebyshev nodes as interpolating points in MATLAB.</li> </ul> </article> </main></body></html>