StudySprint

StudySprint - AI flashcard generator. Upload PDFs/notes → Get smart flashcards in seconds with spaced repetition. Like Anki, but automated. Free for students. Study 3x faster.

Comment

StudySprint: From Exam Anxiety to AI-Powered Learning

🎯 Inspiration

As a student, I've spent countless hours creating flashcards manually—typing out questions, copying definitions, formatting everything perfectly. Then I discovered Anki and loved its spaced repetition algorithm, but there was still one massive problem: I was spending more time making flashcards than actually studying.

During exam season, I'd look at a 50-page PDF textbook chapter and think: "This will take me 3 hours to turn into flashcards." That's when the idea hit me—what if AI could do this in seconds?

I wanted to build something that combined: • The instant convenience of modern AI (like ChatGPT) • The proven effectiveness of spaced repetition (like Anki) • A beautiful, distraction-free interface that makes studying enjoyable

StudySprint was born from the simple belief: students should spend their time learning, not formatting flashcards.

🛠️ How I Built It

Technology Stack

Frontend & Framework: • Next.js 16 with React 19 for server-side rendering and optimal performance • TypeScript for type safety and better developer experience • Tailwind CSS 4 for rapid, responsive UI development • Framer Motion for smooth animations and transitions

Backend & Data: • Prisma ORM with SQLite for local-first database management • API Routes for serverless backend functionality • pdfjs-dist for client-side PDF text extraction

AI & Intelligence: • GPT-4o via Puter.js for free, unlimited AI-powered flashcard generation • Custom prompt engineering to detect content format (quiz vs. notes) • Intelligent text preprocessing for messy PDF extractions

Architecture Decisions

Why Next.js? The App Router gave me powerful server-side capabilities while maintaining a snappy client experience. API routes made it easy to separate concerns.

Why SQLite? For an MVP, SQLite offers zero-configuration persistence. Students can use it locally without any cloud dependencies. Migration to PostgreSQL is trivial if needed later.

Why Puter.js? This was a game-changer—free, unlimited access to GPT-4o without API key management. Perfect for a student-focused free tool.

🚀 The Building Process

Phase 1: Core Flashcard Generation (Days 1-2)

Started with the most critical feature: PDF → Flashcards pipeline typescript The AI prompt engineering was crucial. I created two distinct detection paths:

Format A (Quiz/Q&A): Preserve original questions, extract clean answers Format B (Raw Notes): Generate new flashcards using cloze deletion and Q&A

Phase 2: Spaced Repetition Algorithm (Day 3)

Implemented a simplified version of the SM-2 algorithm:

$$ \text{Next Interval} = \begin{cases} 1 \text{ min} & \text{if Hard} \ 2 \times \text{previous} & \text{if Okay} \ 3.5 \times \text{previous} & \text{if Got it} \end{cases} $$ typescript This ensures optimal retention curves while remaining simple to implement.

Phase 3: UI/UX Polish (Days 4-5)

Built a fullscreen Anki-style review interface with: • Keyboard shortcuts (Space to reveal, 1/2/3 for ratings) • Smooth card flip animations • Progress tracking with visual indicators • Color-coded difficulty buttons (red/yellow/green)

💪 Challenges Faced

Challenge 1: Messy PDF Text Extraction

Problem: PDFs extracted text like: Solution: Built a multi-layer preprocessing pipeline:

  1. Remove page markers during extraction
  2. Preserve line breaks using hasEOL property
  3. Clean up excessive whitespace
  4. Filter out artifact lines

The AI prompt also includes explicit instructions to ignore page numbers, headers, and focus on substantive content.

Challenge 2: AI Generating Poor Flashcards

Problem: Initial attempts produced: • Cloze cards with blanks on articles: "Chemical reactions occur in _____." / "an" • Deck names from question fragments: "Which company recently became" • Broken sentences as flashcards

Solution: Extensive prompt engineering with: • Concrete examples of good vs. bad flashcards • Explicit rules: "Replace KEY TERM, NOT articles or filler words" • Format detection to handle quiz content differently than raw notes • Fallback deck name generation with artifact filtering

Challenge 3: Database Connection Issues

Problem: Each API route created a new PrismaClient(), causing connection exhaustion and data persistence issues.

Solution: Implemented the singleton pattern: typescript Challenge 4: Cascade Deletes Not Working

Problem: Deleting a deck didn't remove associated cards and review logs, causing orphaned data.

Solution: Added onDelete: Cascade to Prisma schema: prisma

📚 What I Learned

Technical Skills

  1. Advanced Prompt Engineering: Learned to guide LLMs with explicit examples, format detection, and edge case handling
  2. PDF Processing: Understanding text extraction quirks and implementing robust preprocessing
  3. Database Design: Proper foreign key constraints, cascade rules, and connection pooling
  4. Next.js 16 App Router: Server components, API routes, and client-side interactivity patterns

Soft Skills

  1. User-Centric Design: Students need simplicity—no config, no signup, just upload and study
  2. Iterative Development: Started with broken AI outputs, refined through testing with real content
  3. Problem Decomposition: Breaking complex features (AI generation) into testable, debuggable steps

Key Insights

LLMs need structure: Vague prompts produce vague results. Explicit examples and rules are essential.

Edge cases matter: Messy PDFs, quiz formats, broken text—handling these makes or breaks the UX.

Local-first is powerful: SQLite + Prisma gives students full control without cloud dependencies.

🎨 Design Decisions

Why Fullscreen Review Mode?

Research shows distraction-free environments improve retention by up to 40%. The fullscreen mode removes navigation, sidebars, and distractions.

Why Auto-Generated Deck Names?

Students shouldn't have to think of creative names. The AI analyzes chapter titles, main topics, and content themes to suggest 3-5 relevant options.

Why No Authentication (Yet)?

For an MVP, removing signup friction means students can try it in 30 seconds. Future versions can add accounts for cross-device sync.

🚀 Future Enhancements

• Image Occlusion: Support for diagram-based learning • Audio Cards: Pronunciation practice for language learners • Anki Export: Let users export to .apkg format • Collaborative Decks: Share flashcards with study groups • Mobile Apps: Native iOS/Android for on-the-go studying

🏆 Impact & Metrics

Built in 5 days, StudySprint demonstrates: • Instant value: 20 flashcards in <10 seconds • Zero friction: No signup, no API keys for users • Science-backed: Spaced repetition increases retention by 40-60%

Students can now transform a 50-page textbook chapter into a complete study deck in under a minute. That's the power of AI meeting proven learning science.

Made with ❤️ for students who want to study smarter, not harder.

Built With

Share this project:

Updates