StudyHub — Peer Academic Collaboration Platform

Status: Deployed and running. Core systems — auth, feed, threads, homework, AI assistant, real-time presence — are complete and functional. Known gaps and planned improvements are documented rather than hidden.

Live: https://studyhub-two-psi.vercel.app/

Try this:

• Sign up and complete onboarding to see compatibility-ranked matches
• Post a question, then use the post AI action menu to ask questions about the post
• Create a thread and @mention Learnora to see the AI reply with thread context
• Check the leaderboard, then look at your own rank card — the ranked list is cached, your row never is

StudyHub is a peer-to-peer academic platform for a university student body, built around one idea: reputation earned by helping other students should be the platform's actual currency, not a vanity number bolted onto a forum. It combines a Q&A/discussion feed, a connections-based social graph, private group chat (Threads), a homework help marketplace, and a multi-provider AI study assistant ("Learnora") into a single system where almost every feature either produces reputation, consumes it as a signal, or reinforces the behavior that generates it.

A student stuck on a problem set at 11pm has no reliable way to find a classmate who already understands the material, is online right now, and is willing to help — outside of scattered group chats with no structure, no accountability, and no way to reward the people who actually show up. StudyHub's bet is that peer tutoring is abundant but undiscoverable: every class has students who are strong in a subject and students who need help in that same subject, but there's no matching layer and no reputation signal for who's reliable. The product surfaces the right people to connect with (department, subject overlap, complementary skills, mutual connections), gives every helpful action a transparent point value, and wraps embedded AI tutoring directly around the connections graph instead of leaving students to coordinate over unrelated tools.

The backend is a Flask monolith on PostgreSQL (SQLAlchemy), with real-time features over Socket.IO, Cloudinary for media, Flask-Mail for transactional email, Firebase Cloud Messaging for push, and a self-built multi-provider AI layer (Gemini, Groq, Cohere, Cloudflare Workers AI, Mistral, OpenRouter) with automatic classified failover. Redis backs four structurally different roles across the app — AI provider state, WebSocket presence, distributed job locking, and durable background job queues — following a horizontal-scaling pass that moved every piece of previously single-process state onto Redis (see Status & Recent Changes). The frontend is vanilla JavaScript (no framework) — a deliberate choice to keep the client lightweight and to force the backend API to be genuinely self-sufficient.

---

Why I Built This

Every university has the same quiet inefficiency: the person who could explain a concept in two minutes is somewhere in the same building, and there's no way to find them. Group chats are ephemeral and low-signal. Nobody gets credit for being the person who always answers. I built StudyHub to see what a campus looks like when helping is legible — when a marked-helpful answer or a login streak rolls up into something visible and comparable, and when an AI assistant is embedded directly in the places students are already working instead of living on a separate chatbot page.

I also used this project to go deep on things that don't show up in a typical CRUD app: a multi-provider AI layer resilient enough that a rate-limited key or a down provider degrades the feature instead of breaking it; a service/route architectural boundary enforced by CI rather than code-review convention; a reputation system with exactly one write path, after inheriting a version of the codebase where the same tier table existed in four places and quietly disagreed with itself at the boundaries; and, more recently, taking the app from a single-process assumption to something that runs correctly under multiple Gunicorn workers — closing the gap between "works locally" and "safe to actually scale horizontally."

---

This is the beginning of the README. Read the rest at the original. Read the full README ↗