I built an adaptive learning platform with Elo ratings as a solo BTech student.Here's what I learned

G S Prajwal — Sat, 21 Mar 2026 15:12:59 +0000

Over the past few months I've been building LearnPathAI
completely solo while managing college. Here's the core idea
and the hard lessons from building it.

The Problem

Every learning roadmap online is identical. Whether you're a
complete beginner or someone who's been coding for 2 years,
you get the same path. That never made sense to me.

The Idea

What if your roadmap adapted to what you actually know?

LearnPathAI parses your GitHub profile or resume to infer
your starting skill weights, then builds a personalized skill
graph around your real gaps not a generic checklist.

As you take quizzes, your proficiency per skill updates using
an Elo rating system (same model used in chess). The roadmap
adjusts in real time skipping what you've mastered,
surfacing what you need next.

Hard lessons building it

Cold start is the hardest problem. Nobody wants a 30
question assessment upfront. GitHub/resume parsing to infer
starting skill weights was the solution — it gives the system
enough signal to bootstrap a useful path without asking
anything.

Elo works surprisingly well for skill rating. But
confidence scores overflow easily if you're not careful with
the math. Spent way too long debugging that.

Dependency ordering in a skill graph matters more than
difficulty levels. Getting the edge direction right between
concepts is what makes the roadmap actually make sense to
follow.

Deploying solo is humbling. Rate limiter bugs from shared
reverse proxy IPs, OAuth issues in production that don't
exist locally, Render cold starts, all hit me in the same
week.

The stack

React + Vite on Vercel, FastAPI on Render, Supabase
PostgreSQL, Clerk for auth, Gemini for quiz generation.

Try it

It's rough around the edges and some flows are still being
built, but the core loop works.

[Link]
https://adaptive-learning-system-with-integ.vercel.app

(First load is slow on Render free tier, please be patient 😭)