M.S. Thesis · CSULB CECS · 2026

Dipak Yadav

Official thesis page for Dipak Yadav, a Nepali software engineer and M.S. Computer Science student at California State University, Long Beach (USA). Path: Nepal → Bangalore, India → Long Beach, USA — author of the UML Generation Pipeline thesis application.

Automated UML Dataset Generation from Natural-Language Requirements with Multimodal Verification

Open live thesis app ↗ Portfolio Google Scholar ↗

Author Dipak Yadav · dipak5501.github.io

Degree M.S. Computer Science · California State University, Long Beach (CSULB) · Long Beach, USA · Jan 2025 – Dec 2026

Prior degree B.E. Computer Science & Engineering · New Horizon College of Engineering · Bangalore, India · 2018–2022 · GPA 8.42/10 · First Class with Distinction

Path Nepali (Nepal) → Bangalore, India → Long Beach, USA

Committee Dr. Yutong Zhao (Chair); Dr. Muhammad Abdul Basit Ur Rahim; Dr. Wei Wei

Application Live UML Generation Pipeline

Profiles LinkedIn · GitHub · Scholar

What Dipak Yadav built

This thesis by Dipak Yadav asks whether automated checks can reject UML diagrams that look finished but are wrong. The running application turns natural-language requirements or source code (Java, Python, C) into design-phase UML (class, object, component, package), renders black-and-white PlantUML images, and scores them with a multimodal VLM ensemble.

Three-stage pipeline: tech spec → PlantUML → render + VLM scoring
Composite score S and majority-vote gate A
FastAPI + Streamlit thesis application with analytics and export
Fine-tuned PlantUML LoRA on Apple Silicon for code → UML stages

The live demo URL can change when the public tunnel restarts. Prefer this thesis page for a stable link; use Open live thesis app for the current running demo.