🏆 3rd Place Winner – Anthropic AI Hackathon @ UofT (Nov 2025)

Track(s): 🟦 Track 1 – Reasoning Systems (UofT AI) 🟩 General Track – Open Innovation

QED is a multi-agent study coach powered by Claude that trains students to think through hard problems instead of just handing them the answer.

Students can:

• Break down tough questions into smaller steps
• Get guided, Socratic help instead of full solutions
• Receive structured feedback on their own work
• Generate realistic study plans for upcoming exams

This project was built for the Anthropic AI Hackathon @ UofT (Nov 1–23, 2025).

Paste a problem (math / CS / econ / theory / etc.) and QED will:

• Decompose it into a sequence of reasoning steps
• Guide you through each step with questions and hints
• Adapt to your answers (gives more hints if you’re stuck)
• Only reveals the full solution outline after sufficient effort

Goal: build your reasoning muscles, not replace them.

Paste the problem and your attempted solution.

The system:

• Checks for logical gaps and unjustified steps
• Highlights missing edge cases or incorrect assumptions
• Rewrites your solution in plain language so you can see if it matches what you meant
• Provides structured feedback: “What you did well / What to improve”

Give QED:

• Course name (e.g. “CSC458 – Computer Networks”)
• Topics or a rough syllabus
• Exam date & weekly study hours

It will:

• Build a realistic day-by-day study plan
• Emphasize high-value topics and spaced review
• Generate checkpoint questions for each topic so you can self-test

Internally, QED uses specialized Claude “agents” implemented as separate prompt profiles:

• 🧩 Decomposer Agent – breaks problems into steps & required concepts
• 🗣️ Socratic Coach Agent – interacts with the student step-by-step
• 🔍 Critic / Verifier Agent – evaluates solutions and explains issues
• 📅 Planner Agent – turns topics + constraints into a study schedule
• (Optional) 🧠 Misconception Tracker – surfaces recurring patterns of mistakes

The frontend orchestrates these agents via a simple backend API, so each mode has a clear contract (inputs / outputs) but shares context when needed.

QED is explicitly designed to support learning, not cheating.

We implement several guardrails:

• No direct full solution by default – the coach uses hints and questions first
• “Show solution” is gated – only appears after multiple attempts or user confirmation
• Clear disclaimer: do not paste take-home exams; use for practice & understanding
• Prompts encourage reflection: after solving, students are asked what they learned and what to do differently next time

This aligns with the hackathon’s focus on safe, human-centered AI and responsible model use.

• Frontend: Next.js (React + TypeScript), Tailwind CSS
• Backend: Next.js API routes / Node.js + Flask (Manim visualization service)
• LLM: Anthropic Claude API (or OpenAI)
• Visualization: Manim Community Edition
• Storage (optional): SQLite / Supabase / PostgreSQL (for saving sessions & history)

You can swap in your own stack; the core idea is agent-like prompt separation.

Ready to deploy QED to the internet? We've got you covered!

Follow the Quick Start Guide to deploy to Railway in minutes.

See Internet Deployment Guide for:

• Railway (recommended)
• Vercel + Railway
• Render
• Custom VPS deployment
• Cost comparisons and monitoring

See Deployment Guide for Docker Compose and local deployment.