Software Engineer | LIU CS, May 2026
I am a Computer Science graduate from LIU Honors, Class of 2026. My focus is on building AI-powered systems for regulated industries. At NYC Emergency Management's Office of the Chief Counsel, I built LegalDocuMan, an LLM-assisted document classification system deployed to process 16,000+ contracts.
Outside of work, I extend my machine learning fundamentals through projects like a custom autograd engine with tensor operations and a financial forecasting pipeline with honest baseline comparisons.
I am interested in roles where engineering substance and AI/ML depth both matter: full-stack work, AI infrastructure, document automation, and regulated-industry tooling.
Designed and built a custom event registration and tracking website (HTML, CSS, JavaScript) integrated with the Zoom Events REST API, handling sign-up and attendance tracking for 1,000+ registrants across hybrid events and replacing a manual coordination process with a reusable template adopted by the department.
Authored a vendor evaluation framework benchmarking 8 competing legal matter management platforms (Thomson Reuters-class systems) on technical capabilities, cost, and integration fit; framework adopted by Chief Counsel and drove the department’s final vendor selection.
Designed and deployed an end-to-end OKR tracking and reporting system for the Office of the Chief Counsel: built a dynamic Power Apps form with cascading dropdowns backed by a SharePoint list of 63 pre-populated records across 4 attorneys, and orchestrated 3 Power Automate flows for quarterly reminders, 48-hour compliance follow-ups, and automated HTML report generation and email delivery.
Re-implemented a reverse-mode automatic differentiation engine in Python (inspired by Karpathy’s micrograd), then extended it beyond the tutorial with tensor operations, a numerical gradient checker using finite differences, and a PyTorch benchmark suite validating gradient correctness against torch.autograd on identical inputs.
Built a character-level language model on top of the custom engine with tokenization, embedding layers, and an MLP architecture following Bengio et al. (2003), trained end-to-end via backpropagation on a 32,000-name dataset.
Built a modular Python application for automated legal document classification, signature detection, and vendor-based file organization — deployed at NYC Emergency Management’s Office of the Chief Counsel to process a legal drive of 16,000+ contracts ahead of migration to a new legal management system.
Architected as a multi-component pipeline (processing engine, threaded GUI, CLI query tool, test suite) with OCR fallback via Tesseract, PDF text extraction via pdfplumber, and persistent metadata tracking for retention and destruction scheduling; released as MIT-licensed open-source with full installation and contribution documentation.
Python pipeline for next-day price prediction on AAPL using 9 technical indicators. Predicted next-day returns to handle non-stationarity, then reconstructed price for evaluation against a naive baseline on a held-out test set.
XGBoost achieved 1.54% MAPE vs. 1.05% for the baseline, a known finance-ML result.
Built a sanitized public version of the Disaster Law Symposium registration and tracking website used for hybrid event sign-up and attendance workflows.
Implemented reusable front-end registration flows with HTML, CSS, and JavaScript, designed around the same operational needs as the internal Zoom Events-integrated system.
I am happy to discuss software engineering, cybersecurity, research opportunities, and professional collaborations. Feel free to reach out with any inquiries or potential opportunities.