What I have shipped.

A desktop app that locates any file from a plain-English description, with no need to remember the filename or folder. Sole architect and developer.

Hybrid retrieval: Lucene.NET BM25 keyword search and on-device ONNX vector embeddings, blended with reciprocal rank fusion. Everything runs on-device, so document content never leaves the user's machine. That is the core privacy differentiator for enterprise.

Full-content indexing across PDF, DOCX, PPTX, XLSX, HTML, Markdown, text and email, with live file-system watching for incremental updates. Production-ready, now being tailored for enterprise deployment alongside company OneDrive.

Built and integrated a fraud-reporting automation module into the Beyon Admin Portal, replacing a manual six-step process with a two-click system. Java Spring Boot, Maven, Apache POI / Docx4J, Thymeleaf, Docker, and AWS (EC2, S3, IAM).

Reduced fraud-case processing from 15 to 30 minutes down to under 10 seconds, a 98% reduction and roughly 800 administrative hours saved a year. Secure by design: role-based access control, audit logging, presigned S3 URLs, HTTPS-only, secrets managed through CI.

Still in daily production use by the Fraud & Compliance team after the internship ended. They reported faster turnaround, less after-hours backlog, and more accurate reports.

Led the team as Scrum Master to build a student-affairs management system under hackathon time pressure. Python (Flask), JavaScript, HTML/CSS, SQLite.

Automated a seven-step manual workflow, cutting approval time from one to two weeks down to three days and saving over 1,000 administrative hours a year. Staff throughput went from about 50 forms over several weeks to 75 a week.

A centralized admin dashboard with real-time tracking, filtering, and one-click approvals cut administrative workload by more than 70% and retired the physical document archive.