Ranuj Vinson O. Soriano

Computer Science student, expected to graduate in 2027

Ranuj Vinson O. Soriano

I turn ideas into working software, then test and refine it until it behaves the way I intended.

View selected workHow I build

Selected work

  • Group thesis project

    CALSS

    Windows app that flags code similarity and coding-style anomalies in student C submissions for instructor review.

    Illustration only, not a screenshot

    • CALSS flagged-pair panel for two synthetic students, Student_A and Student_B, labelled “Exact Duplicate” with 99.2% combined, token 100% and style 98%. Student_A’s exam is rated 78.1%, Moderate Likelihood, with 13 of 14 features matching their profile; Student_B’s is 53.7%, Low Likelihood, with mismatches in bracket placement, blank line usage and operator spacing. The AI evidence note says AI narration was not requested.
    • CALSS source comparison viewer showing a synthetic midterm exam for Student_B, with File A and File B tabs. A banner reads “17 matching lines · 1 section breaks this student’s pattern”. Most of the code is highlighted gold, and the main function is labelled “braces on same line (usually next line) · operators spaced (usually unspaced)”.
    • CALSS Classes tab listing five synthetic students, Student_A to Student_E, each with three submissions, a style chart and habit notes such as “never writes comments”, “next-line braces” and “heavy use of single-letter variables”. Student_A and Student_B are each flagged once, with a maximum of 99.2%.
    • CALSS overview after analysing 20 synthetic files from 5 students: 1 of 10 pairs flagged for review, labelled exact duplicate; maximum score 99.2%, average score 67.8%; the AI instructor summary reads “AI summary not requested”.

    Group thesis project

    • Flagged pair breakdownCALSS’s “Exact Duplicate” category at 99.2% combined. A flag is a prompt for review, not proof.
    • Source comparison viewerStudent_B’s exam in the File B tab. Gold lines also appear in Student_A’s file, shown in the File A tab.
    • Class style profilesCoding-style habits for each student, built from three activities each.
    • Analysis overviewOne of ten exam pairs flagged for review, with the AI summary off.

    Genuine screenshots of CALSS 1.3.1 · synthetic data, no student work

    Open the CALSS case study

  • Personal project

    Jairus

    Experimental Python voice assistant for playing music by voice. On hold: in its latest test, playback didn’t work.

    Illustration only, not a screenshot

    Personal project

    Illustration only, not a screenshot or recording. Development is on hold; in the latest test, playback didn’t work.

    Open the Jairus case study

Two projects I'm proudest of. Each opens a case study with its evidence and current limitations.

More projects will be added once their current state and evidence have been checked.

How I build

AI tools speed up how I implement and explore. They don’t replace understanding what I’m building.

Where AI tools help

Drafting code and tests, exploring libraries and approaches, and reviewing designs for problems I missed.

What stays my responsibility

Requirements, architecture and technical decisions, testing and debugging real behavior, security and privacy, and honest claims about the result.

The workflow

The pattern I follow now, not a record of every past project.

  1. Define the problem

    What it must do, what it must not do, and how I’ll know it works.

  2. Research and plan

    Compare approaches and settle the architecture before writing code.

  3. Implement

    AI assistants such as Claude Code often draft much of the code, under my direction.

  4. Review, debug and test

    Run the real program and the tests, and trace each failure to its cause.

  5. Refine on evidence

    Fix what the results show and record what still doesn’t work.

In practice

  • CALSS: reproducible and measured

    A golden test requires byte-identical output on three synthetic datasets, and caching cut a 2,973-file run from about 28 s to 7.4 s on the development PC. That shows consistency and speed, not detection accuracy. CALSS testing

  • Jairus: gateway, not the whole assistant

    Each of 11 phone-gateway security fixes was undone in turn, and the tests failed every time. That checks the gateway only; the end-to-end voice workflow has no automated tests. Jairus testing

About

Make it work. Understand why.

I’m Ranuj Vinson O. Soriano, a Computer Science student expecting to graduate in 2027 and working toward an entry‑level software development role.

What keeps me building is seeing an idea turn into software that actually runs, then the slower work of making it behave the way I meant. So far that has meant Windows desktop tools, a voice assistant, phone-to-PC utilities and small automation scripts. Two are documented in depth: CALSS, my group’s thesis tool that flags possible code copying for an instructor to review, and Jairus, an unfinished personal voice assistant for playing music, now on hold.

The part I find most interesting is what happens after the first version runs: testing what it really does, tracing failures to their cause and fixing them properly. My case studies list what doesn’t work yet alongside what does.

Where I want to grow

I haven’t settled on one specialty yet. These are the areas I’m working toward.

  • Backend development
  • Testing and automation
  • Applied AI
  • Computer vision
  • Desktop and cross-device tools

Contact

Here’s where to reach me.