Entry-Level Data Analyst

Eliminated 1.5 hours of daily manual data collection by automating weather ingestion from a public API every 10 minutes into SQLite across 5 cities, using Python and GitHub Actions for scheduling and monitoring.

Reduced fare and demand analysis from 7 hours to 15 minutes across 20M+ NYC taxi trips by ingesting 8 months of Parquet files into a normalized PostgreSQL schema, building a Flask REST API, and deploying an interactive React dashboard across 256 taxi zones.

Enabled real-time exploration of wage disparities across 344K records spanning 1981 to 2013, reducing manual analysis from 1.5 hours to 15 minutes across 6 demographic dimensions including age, education, and occupation, using an interactive D3.js scrollytelling interface.

Conducted a comparative study of traditional and graph-based machine learning models for detecting illicit cryptocurrency transactions, achieving up to 0.97 accuracy and 0.88 macro F1 on the Elliptic dataset and 0.9857 accuracy with 0.9312 macro F1 on the Ethereum dataset.

Trained and evaluated NLP models on the ISOT Fake News Dataset (44,898 articles), achieving up to 100% accuracy, precision, recall, and F1-score using RoBERTa for real versus fake news classification.

Evaluated empirical privacy risks of Personalized Differential Privacy using membership inference attacks, achieving up to 98.59% test accuracy on MNIST while observing AUC values closely aligned with individualized privacy budgets across MNIST, CIFAR-10, and SVHN.

Built a data-centric object detection pipeline for camouflaged animals, achieving up to 69.78% mAP@0.5 on real-world wildlife data despite extreme foreground–background similarity.

Evaluated a Perceiver model trained on compressed image embeddings instead of raw pixels, achieving up to 94.4% classification accuracy while reducing the computational cost of vision transformers.

Trained and compared DCGAN, StyleGAN, and VAE models on approximately 15,000 textile images across six pattern categories, generating designs at 64×64 and 256×256 resolutions for automated pattern synthesis.

Built a full-stack pharmacy management system implementing 4 role-based workflows, 8 REST API modules, and 5 MongoDB data models using dual Angular frontends and a Node.js backend.

Built a decision tree classifier from scratch to predict cardiac arrest risk, achieving 81.11% accuracy on real clinical data with an interactive diagnostic interface.