LevelUp Pro — AI-Graded Hands-On Tech Labs

Role Simulation lets you step inside real jobs - Developer, QA, Support etc(31) and do the actual work in realistic environments, with AI evaluating your performance before you choose a right career.

Comment

LevelUp Pro — Experience the Job Before You Choose It

💡 Inspiration

Most people pick a career on guesswork. They watch tutorials, finish courses, collect certificates — and still have no idea what the job actually feels like, or whether they'd be any good at it. After 30 years in enterprise IT, I watched capable people wash out of roles they'd never truly tried, and watched hiring managers struggle to tell "completed a course" apart from "can actually do the work."

The gap isn't content — there's infinite content. The gap is proof of performance: a way to do the real work of a role and get honest, senior-level feedback on how you did, before committing years to a path.

That's what LevelUp Pro is built to close.

🎯 What it does

LevelUp Pro is Role Simulation — and it runs in two tracks.

Guided labs walk a learner through a scenario with step-by-step support, ending in a quiz and a certificate. This is the on-ramp.

Unguided labs (SimOps) are the real test — a 6-stage assessment that simulates real enterprise incident response, with no hand-holding:

  1. Pre-Lab — sequence the correct response procedure
  2. Role Lab — a simulated terminal where you work on task or fix a live incident or complete a Dev task.
  3. Quiz 1 (Process) and 4. Quiz 2 (Domain) — process and domain knowledge
  4. Mock Interview (Process) and 6. Mock Interview (Domain) — open-ended scenario questions

The unguided track is where the AI does its most important work — judging open-ended solutions with no answer key to match against. Every stage feeds a proctored, weighted composite score, and a learner must reach ≥ 90% with all stages complete to earn certification — a real bar that people genuinely fail. Clear it, and you earn a downloadable certificate with a unique ID, a LinkedIn-shareable credential, and a public verify URL.

🤖 Where AI makes the call

The key judgment — did this person perform like a professional? — is made by Google Gemini 2.5 Flash, live in production. For e.g.

  • In the Incident Lab, Gemini evaluates each terminal command you run against a real scenario — a crashed Node.js/PM2 app, an exhausted PostgreSQL connection pool, a DNS resolution failure, NTP clock drift breaking JWTs, a TCP SYN flood — and returns pass / fail / pending-review with reasoning.
  • In the two Mock Interviews, Gemini reads your free-text answers to scenario questions (e.g. "Walk me through your exact first steps in the first 5 minutes of a P1 outage") and scores each on relevance, communication, and structure, with written feedback on what a strong answer covers.

No human reviews these. Gemini grades every submission in seconds, feeding the weighted composite that gates certification.

🏗️ How I built it

A TypeScript monorepo where one Node.js process serves both the API and the React SPA:

  • Frontend: React 18 + TypeScript, Wouter, TanStack Query, shadcn/ui + Radix, Tailwind, Framer Motion, Recharts, Vite 7
  • Backend: Node.js 20 + Express, Passport.js (Google/GitHub/Yahoo/LinkedIn OAuth), Firebase phone OTP, express-session
  • AI evaluation: Google Gemini 2.5 Flash via generativelanguage.googleapis.com (server/gemini.ts)
  • Data: PostgreSQL + Drizzle ORM for users/commerce/certificates; Firebase Firestore for high-frequency SimOps lab data
  • Payments: Razorpay (primary, INR/USD, HMAC SHA-256 verification) + Stripe
  • Proctoring: tab-switch and copy-paste detection logged to Firestore, with score penalties

Humans design each role scenario once. Everything at the point of learning — grading commands and answers, generating feedback, computing composites, enforcing proctoring — runs on Gemini and Google Cloud. That's why one founder serves learners across India, Indonesia, and South Africa with no grading team.

📚 What I learned

  • AI grading only earns trust when it's specific. Vague feedback got ignored. Scoring on named dimensions with a concrete "here's what a strong answer covers" line is what made it feel like a senior engineer, not a black box.
  • Weighting is a product decision, not a formula. Making the hands-on lab worth 40% and requiring 100% to advance shapes learner behaviour more than any content.
  • Integrity matters the moment scores mean something. Once a composite gates a certificate, people try to game it — so proctoring and violation penalties had to be real.

🧗 Challenges I faced

  • Consistent AI grading. Open-ended answers are hard to score reproducibly; getting stable, fair Gemini scores took real prompt and rubric engineering — plus a graceful fallback (pending-review, never auto-fail) when a grading call errors.
  • An honest composite. Per-stage rounding was needed so the displayed total always matches the weighted-contributions table students can verify by hand.
  • Solo, end to end. Design, build, AI integration, payments, proctoring, and going live — all as one founder.

💰 Real product, real users, real revenue

LevelUp Pro is live today, with real learners and paid transactions through Razorpay. Pricing starts at ₹399 for an individual lab — deliberately low, because for a first-generation jobseeker, being able to try the role before committing years to it is a fundamentally different kind of access.

🌍 Why it matters — Education & Human Potential

LevelUp Pro changes how people learn, grow, and reach their best work: they stop guessing what a job feels like, get senior-level feedback on real performance, and can prove they've already done the work — before they ever apply for it.

Built With

Share this project:

Updates