Learnspree

Homepage
Scheduling
Chat
Agent

Inspiration

We've all been there—staring at a textbook at 2 AM, desperately trying to absorb information before an exam. Traditional studying is lonely, unstructured, and frankly... boring.

While AI chatbots have revolutionized many fields, studying with them feels like talking to a wall of text. They answer your question, but they won't challenge your understanding, quiz you until concepts stick, or motivate you when you're burned out.

We asked ourselves: What if studying felt less like a solo grind and more like being in a study group—with really smart friends who never get tired?

That's how Learnspree was born.

What it does

Learnspree is an AI-powered study platform featuring multiple AI agent personas that work together to help you master any topic:

🧠 Study Marathon Agent — Orchestrates your entire preparation journey, evolves as you learn
🎯 Professor Quiz — Tests your knowledge with adaptive quizzes
📚 Dr. Clarity — Explains concepts in crystal-clear language
⚔️ The Challenger — Asks devil's advocate questions to deepen understanding
💪 Coach Spark — Keeps you motivated when focus fades

Key Features:

Intelligent Agent Handoffs: Agents seamlessly pass conversations based on your learning state
Mastery Tracking: Visual progress from Beginner → Practicing → Proficient → Mastered
Quiz Mode: Answer questions correctly to unlock deeper challenges
Smart Scheduling: Say "Plan 5 days for my React exam" and get a structured study timeline

How we built it

Frontend: React 19 with Vite for blazing-fast development, styled with TailwindCSS

Backend: Node.js with Express 5, using Mongoose for MongoDB Atlas integration

AI: Google Gemini 2.5 Flash API powers all agent interactions, with thinking capabilities for complex reasoning

Architecture: We implemented a persona-based agent system where each agent inherits from a BaseAgent class, has unique personality prompts, and can trigger handoffs to other agents based on conversation context.

Challenges we ran into

Agent Handoff Coherence — Getting agents to smoothly hand off without losing conversation context. Solved by maintaining session history across all agents with a shared MongoDB model.
Quiz Answer Detection — Distinguishing "yes" (ready for quiz) vs "yes" (as an answer). We built careful heuristics to detect actual responses vs conversational acknowledgments.
API Rate Limits — Gemini occasionally returns 503 during high load. We learned that simple error handling with clear user messaging beats complex silent retry logic.
Real-time Mastery Updates — Ensuring the progress bar updates after quiz feedback required careful React state management with useCallback patterns.

Accomplishments that we're proud of

The agent personality system — Users genuinely feel like they're talking to different tutors, not a single bot
Mastery tracking — Watching that progress bar fill up is genuinely satisfying and keeps users engaged
Seamless handoffs — Agents transition naturally with visible reasoning, making the experience feel cohesive
Sub-second response times — Using Gemini 2.5 Flash-Lite for standard queries keeps the experience snappy
Quiz-to-Challenge pipeline — The automatic escalation from basic quizzes to deep challenges creates a real learning journey

What we learned

Multi-agent systems need clear role boundaries — Vague personas lead to overlapping behavior
User feedback loops accelerate learning — The mastery bar provides dopamine hits that keep users engaged
AI personality matters — Users responded better to "Professor Quiz" than "Quiz Agent"
Graceful degradation > complex retry logic — Simple error handling beats silent retries
Vite + React 19 is incredibly fast — Hot reload under 100ms made iteration a joy

What's next for Learnspree

-Voice Mode — Talk to your study squad hands-free -Collaborative Sessions — Study with friends + AI agents together in real-time -Spaced Repetition — Intelligent review scheduling based on forgetting curves