Ezekiel Galauran

01 — About

02 — Selected work

• Client system

  In development for deployment

  Dorotheo Dental Clinic Management System

  A clinic management system for a dental practice, built by a team of 5 developers.

  • Django
  • Django REST Framework
  • Next.js
  • TypeScript
  • PostgreSQL

  Role

  Project Manager & Full-Stack Developer

  Scale

  • Team of 5 developers
  • 4 user roles
  • 17 core workflows
  • 20+ protected API surfaces

  What I built

  • JWT authentication
  • Role-based access control
  • Audit logging
  • Prompt-security controls for the LLM assistant
  • Query and pagination optimization

  The system also includes a RAG/LLM pipeline with intent classification and semantic caching, built by the team.

  Read more about Dorotheo Dental Clinic Management System

• Independent project

  In production use

  Motor Tool & Sub-Component Pumping Hour Tracking System

  An asset-tracking application used by field technicians working in Abu Dhabi, with live company data.

  • Next.js 15
  • React 19
  • TypeScript
  • Prisma
  • Neon Postgres
  • NextAuth
  • Vercel Blob
  • Tailwind CSS
  • Vitest

  Role

  Sole developer

  Highlights

  • Pumping-hour cascade across motors and sub-components
  • Unified 11-state asset status model
  • Assembly/disassembly records
  • File attachments
  • Search-first UI
  • Requirements maintained through PRD versions up to v2.9

  Privacy

  The app holds live company data, so there is no public link or screenshot.

  Read more about Motor Tool & Sub-Component Pumping Hour Tracking System

• Deployment operations

  Deployed · Sept 2026

  NAMFREL Election Portal — Production Deployment

  Main handler of continuous deployment to Vercel for APC's NAMFREL election monitoring portal on election day, Sept 14, 2026.

  • Vercel
  • Git
  • GitHub

  Role

  DevOps & Infrastructure (Sub-Team 4) — Production Deployment

  Context

  APC served as the NAMFREL Election Data Analytics Center for the 2026 BARMM Parliamentary Elections. 46 students and 9 faculty across CS, IT, and CpE built and supported the election analytics system.

  The blocked-deploy problem

  Teammates' merged commits could not trigger production deployments on the team's Vercel plan (shown as "Blocked"), so I shipped their changes by pushing "chore: trigger deployment" commits from my authorized account — 13 production deployments on election day, Sept 14, 2026.

  What I did

  • Main handler of continuous deployment to Vercel on election day (Sept 14)
  • Performed live production redeployments as teammates' changes were merged
  • QA of production deployments during shift, before handoff to the next shift

  The GitHub Actions CI pipeline, server caching, and Playwright automation were built by a teammate.

  • Live portal(opens in a new tab)

  Read more about NAMFREL Election Portal — Production Deployment

• Undergraduate thesis

  AY 2026–2027

  XLM-JUD: Dialect-Aware Smishing Detection

  A study of whether NLLB-200 back-translation improves XLM-RoBERTa smishing detection for Tagalog, Cebuano, and Ilocano text.

  • Python
  • XLM-RoBERTa
  • Meta NLLB-200
  • FLORES-200
  • SVM
  • Logistic Regression (TF-IDF)
  • SMOTE

  Role

  Co-author — research and writing (with three teammates)

  Problem

  Smishing detection in the Philippines is hampered by scarce, imbalanced datasets and Tagalog, Cebuano, and Ilocano text that English-trained filters misread.

  Approach

  Compared original XLM-RoBERTa, XLM-JUD (XLM-RoBERTa trained with NLLB-200 back-translation augmentation), and classical baselines, with and without SMOTE.

  Results (hold-out test set of 11,603 messages)

  • Original XLM-RoBERTa: accuracy 98.61%, Macro-F1 0.9849, MCC 0.9698
  • XLM-JUD: accuracy 98.19%, Macro-F1 0.9802, MCC 0.9606
  • Classical baselines: Macro-F1 0.9373–0.9561
  • Back-translation did not improve classification (210 vs. 161 errors; McNemar's test p < 0.001). SMOTE also lowered classical-model performance.

  Limitations

  • Cebuano and Ilocano test subsets contained only legitimate messages
  • Back-translated data not independently verified
  • Single training run

  Why it matters

  A rigorous negative result — the study reports what the data showed, not what was hoped.

  My contribution was research and writing; I did not write the model code.

  Read more about XLM-JUD: Dialect-Aware Smishing Detection

03 — Experience

1. November 2026 – Present

   Optum — Student Advocate Program

   Intern, Early Careers

   Intern in the Optum Student Advocate Program under the Early Careers team, started November 5, 2026.

2. September 2026

   APC × NAMFREL — 2026 BARMM Parliamentary Elections

   DevOps & Infrastructure (Sub-Team 4) — Production Deployment

   APC served as the NAMFREL Election Data Analytics Center. 46 students and 9 faculty across CS, IT, and CpE built and supported the election analytics system.

   • Main handler of continuous deployment to Vercel on election day (Sept 14)
   • Performed live production redeployments as teammates' changes were merged
   • QA of production deployments during shift, before handoff to the next shift
   • Shipped teammates' changes with "chore: trigger deployment" commits from my authorized account when their merged commits were "Blocked" on the team's Vercel plan — 13 production deployments on Sept 14

   namfrel-barmm2026.apc.edu.ph(opens in a new tab)

04 — Leadership

• Vice President

  APC Band

  Jul 2026 – Present

• Internal Vice President

  APC Microsoft Community (APC-MSC)

  Nov 2025 – Jun 2026

• Executive Director of Events and Programs

  APC Microsoft Community (APC-MSC)

  Nov 2024 – Jun 2025

• Assistant Director of Operations

  JPCS – APC

  Nov 2024 – Jun 2025

Speaking & hosting

• Speaker, Microsoft Sandbox: Building Your Startup with Microsoft 365
• Speaker, SoCIT Officer Workshop 2025
• Host, Microsoft Learn Student Ambassador event

05 — Education

06 — Contact