End-to-end data and AI systems built from scratch — pipelines, models, APIs, and everything in between.
End-to-end fraud detection system processing 284,807 real credit card transactions. Built a PySpark ETL pipeline with behavioral feature engineering, trained Random Forest and XGBoost classifiers, and deployed predictions via a FastAPI REST endpoint containerized with Docker.
NLP pipeline that automatically summarizes earnings call transcripts and analyzes financial sentiment using open source language models. BART for abstractive summarization, FinBERT for domain-specific financial sentiment, and YAKE for keyword extraction. Correctly identified Apple as the only negative report in the test set.
End-to-end experimentation platform simulating an e-commerce checkout button A/B test across 10,000 users over 14 days. Applied z-test for statistical significance, computed confidence intervals and lift metrics, and exposed results via a REST API with on-demand simulation capabilities.
A full-stack mobile and web AI application built around memory, emotional awareness, and continuity — designed to feel fundamentally different from a standard chatbot. Features a semantic memory system, multi-layer emotion detection pipeline, and proactive engagement. Currently in active development.