Hi, I'm Shantnu

Built a full-stack AI-powered receipt parsing and expense splitting application that uses GPT-4o Vision for structured extraction from receipt photos, with confidence scoring and a human-in-the-loop review and correction workflow.

Rebuilt an outdated PHP website (~10-12 years old) for a small courier and delivery business in Alexandria, VA, migrating the entire UI to a modern tech stack with a luxury dark-themed design.

Re-implemented a 28M parameter recurrent language model from scratch, presenting a lighter alternative for step by step reasoning by reusing a compact transformer block.

Engineered a C++ network monitoring system capable of processing 1-5 Gbps packet streams with sub-millisecond p99 latency and <0.1% packet loss through efficient buffer management.

Developed a robust data pipeline to curate "NewsSlant," a corpus of 34,790 English tweets from 48 U.S. news outlets and 4M associated replies, with a strict chronological cap of the first 20 replies per thread (345,182 labeled replies) to mitigate viral outlier bias and manage computational cost.

Built an analytics pipeline for 185K+ NYC 311 requests with thorough data cleaning, outlier removal, and feature engineering.

Developed a full-stack e-commerce bookstore application, serving 40+ book titles across 4 categories, highlighting comprehensive web development skills.

Developed a predictive tool to estimate job salaries, scraping over 900 job listings from Glassdoor using Python and Selenium.

Executed comprehensive data analysis and visualization on a dataset of nearly 8,000 customer entries, employing various classification techniques and comparing their effectiveness using ROC-AUC metrics.