The Problem with Learning Platforms
Most learning platforms are content dumps. Upload a PDF, assign a quiz, check a box. Nobody actually learns anything — they just complete modules. The progress bar fills, the platform reports "engagement," and the student walks away with nothing.
We knew Conesta had to be different.
When we started building, we asked one question: what if the platform understood what you're trying to learn and built the path for you?
That question is the whole product.
AI-Generated Learning Paths
The core of Conesta is its AI learning engine. A student enters a topic — say "distributed systems" — and Conesta doesn't search a library. It generates a personalised roadmap:
- Concept mapping — breaks the topic into prerequisite chains
- Resource matching — pulls from our marketplace of links, videos, notes, and code
- Difficulty calibration — adjusts based on the student's history and pace
We built this on the Claude API. The reasoning was the deciding factor. We needed the model to understand relationships between concepts, not keyword-match its way through a content library and pretend it had a plan.
Real-Time Study Rooms
Learning is social. Conesta's study rooms let students collaborate in real-time, with built-in Pomodoro timers, shared whiteboards, and an ambient lofi mode for focus sessions.
The hard part wasn't the timers. It was syncing state across participants without the interface feeling sluggish. We use WebSocket connections with Redis pub/sub for room state, and optimistic UI updates so the screen reacts the moment you click. If your collaboration tool has a discernible delay, students will switch back to a Google Doc and a Discord call, and you've already lost.
The Stack
- Frontend: Next.js + React with Framer Motion for interactions
- Backend: Node.js API with PostgreSQL
- AI Layer: Claude API for path generation, concept mapping, and content recommendations
- Real-time: WebSockets + Redis
- Hosting: AWS with Vercel for the frontend
Boring choices, deliberately. The interesting part of the product is the intelligence layer. Everything else is plumbing — and good plumbing is the kind nobody notices.
What We Learned
"AI-first" gets used to mean a chat box bolted onto an existing app. That's AI-last with extra steps.
Real AI-first means the model is the foundation. Every design decision in Conesta flows from one question: how does this help the AI help the student? If a feature can't answer that, it doesn't ship.
If you're building an AI product, start with the intelligence layer. The UI is just the interface to it.
Conesta is live at conesta.in. Try it free.