Oscar Manuel Arenas portrait

Oscar Manuel Arenas

Computer Science Graduate Student • Machine Learning • Software Engineer

Welcome 👋

I’m Oscar Arenas, a Computer Science graduate student at California State University, Long Beach. I focus on machine learning and software engineering.

I’m starting my graduate thesis as a Machine Learning Research Assistant with the Data Semantics and Human Data Interaction (D2) Lab at California State University of Long Beach under Dr. Bo Fu . This summer, I finished my internship at the USC Institute for Creative Technologies (ICT) in the Learning Sciences group with Dr. Benjamin Nye and Joel Walsh. I will continue my research with Joel Walsh since he provided a strong foundation for me in statistical analysis and machine learning.

Another highlight was my internship with the Office of Defense Nuclear Nonproliferation with Dr. Paul Adamson, where I developed an ontology visualization tool for the Data Science portfolio to identify gaps in AI research.

Experience

Data Infrastructure & Analysis Intern – Willog

Seoul, South Korea | Dec 2025 – Jan 2026

  • Designed and implemented a production Sales CRM web application
  • Built a Kanban-based lead management system backed by Google Sheets and supporting multi-user workflows
  • Integrated Google Gmail, Calendar, and Drive APIs for email tracking, scheduling, and document linking
  • Developed analytics dashboards with KPIs and interactive charts to visualize pipeline value, win rates, and sales performance
Show Certificate Willog Certificate of Completion – Oscar Arenas

AI Research Intern - USC Institute for Creative Technologies

Los Angeles, California | May 2025 – August 2025

  • Designed a workflow to analyze 3rd to 12th grade English science text datasets to evaluate LLMs such as ChatGPT-4o, Gemini, and locally hosted Hugging Face models on Spanish translations performance using Spanish readability metrics.
  • Developed Python scripts to evaluate untargeted vs. grade-targeted Spanish translations, and implemented a Microsoft AutoGen multi-agent prompt retry system that improved grade-level alignment accuracy by up to 60%
  • Documented workflows and statistical analysis processes so researchers could trace, reproduce, and extend experiments

Data Science Intern – National Nuclear Security Administration

Remote | May 2024 – April 2025

  • Interned at the NNSA Defense Nuclear Nonproliferation R&D office supporting the Data Science portfolio.
  • Created an ontology focused on Trustworthy, Effective, and Deployable AI (TED-AI) using Python to better characterize current focus areas, opportunities, and potential gaps for DNN R&D.
  • Developed a modular VS Code extension with drag-and-drop dataset uploads and D3.js visualizations, enabling non-technical users to explore ontology data.
  • Built an AI agent utilizing the AI ontology to evaluate research documents related to DNN R&D, leveraging Python LangChain libraries to assess ontology effectiveness.

Physics Tutor - Cerritos College

Norwalk, California | 2019 – 2020

  • Provided drop-in tutoring for approximately 5 students per day in introductory physics courses (PHYS 101 and PHYS 201).
  • Guided students through foundational concepts such as kinematics, Newton's laws, energy, and basic thermodynamics.
  • Helped students prepare for exams through problem-solving practice, concept reviews, and worked examples.
  • Adapted explanations to different learning levels and broke down abstract physics concepts into approachable steps.

Programming Instructor - Dreams for Schools

Santa Ana, California | 2023 – 2024

  • Organized lesson plans to teach basic engineering and programming concepts.
  • Taught Arduino fundamentals in C++ and guided students in building a remote-controlled racing car.
  • Led a student team to 1st place in a game development competition using DFS Appmaker - watch video of me.

Accounting Intern - Utopia Transport Inc.

Los Angeles, California | 2019 – 2020

  • Processed approximately 30 invoices daily and managed payroll, budgeting, and financial reporting for a fleet of 15 drivers.
  • Resolved an average of 3 billing disputes per day by coordinating with shipping lines and chassis companies (CMA, Evergreen, DCLI, Trac).
  • Modernized company operations by introducing Microsoft tools and setting up data infrastructure, supporting growth from 5 to 15 drivers.

Projects

Golf Course Segmentation

Python, TensorFlow, Next.js, FastAPI | Live Site

  • Trained and fine-tuned a custom U-Net model (ResNet50 encoder) for 6-class aerial image segmentation of golf course features (fairways, greens, tees, bunkers, water hazards).
  • Built a Next.js web app with Google Maps API integration for dataset creation, annotation, and one-click AI segmentation analysis.
  • Deployed the segmentation model as a FastAPI inference endpoint on HuggingFace Spaces.
  • Implemented end-to-end preprocessing, training, evaluation, and inference workflow.

Translations Annotation Tool

TypeScript, Vite | Live Demo

  • Built a web-based review interface for teachers to evaluate and annotate Spanish translations generated by LLM models as part of the USC ICT research workflow.
  • Developed with TypeScript and Vite for a fast, component-based single-page application with client-side data handling.

🏓 AutoGen Pong (VS Code Extension)

Experimental VS Code Extension

  • Classic Pong game inside a VS Code WebView — paddles controlled by AutoGen agents.
  • Canvas-based frontend with Python + Flask backend.
  • Functional testing setup with Playwright; clean separation between extension frontend and agent backend.
  • Status: environment scaffolded; game logic, backend agent interaction, and Flask routing coming next.

Daily Vending Machine Sales Forecasting

Python, TensorFlow, Pandas | Spring 2025

  • Analyzed time-series sales data from vending machines to identify daily and weekly purchasing trends.
  • Evaluated model performance using RMSE, MAE, and MAPE metrics to assess forecast reliability.

GenInv – Invoice Generator

Personal Project | Test-Driven Development

  • Desktop app to automatically generate invoices using tkcalendar, customtkinter, and PyPDF2.
  • Workflow: select property → select/set default tenant → confirm/adjust next billing date.
  • Supports changing default tenant and auto-advancing billing dates; invoices built from prepared file paths.