How I Built an AI Platform That Verifies Student Skills

#webdev #javascript #ai

The Problem

School students in IT have no real way to prove their skills. Grades show how well you take tests. Certificates show you completed a course. But neither shows what you can actually build.

I saw this problem firsthand. Students with amazing projects couldn't get internships because recruiters had no way to verify their real abilities. So I built Verix.

What is Verix?

Verix is a digital reputation platform where students verify their skills through AI, not paper credentials. It has three main features:

SkillProof AI — Submit a description of your project. The AI analyzes it and scores your skills across 6+ domains: React, Python, ML/AI, Design, Data, and Web Development.

TalentMatch — Swipe through internships and opportunities matched to your verified skills. Think Tinder, but for career opportunities.

EduPath — Get a personalized learning roadmap based on your current level and goals. The AI adapts recommendations as you improve.

How I Built It

The entire platform is a Single Page Application built with HTML, CSS, and vanilla JavaScript. No React. No frameworks. Just clean code.

The AI layer uses Anthropic's Claude API for skill analysis. I designed a scoring algorithm that evaluates projects on originality, complexity, and technical depth. The prompt engineering was the hardest part — making AI evaluations consistent and fair across different domains took dozens of iterations.

Deployment is on Vercel with a mirror on GitHub Pages.

What the AI Actually Does

When a student submits a project for verification, the AI:

Analyzes the technology stack used
Evaluates complexity and originality
Compares it against baseline expectations for that skill level
Returns a score from 0 to 100 with specific feedback

The AI also powers TalentMatch — it matches student profiles against internship requirements to find the best fit.

Challenges

The biggest challenge was making AI scoring fair. A React project and a Python project are fundamentally different. The AI had to understand context, not just match keywords. Prompt engineering took more time than writing the actual UI.

Another challenge: everything runs in a single HTML file. Keeping the code modular and clean without a framework required discipline.