Teachr.me

Save 70% of the time to mark an essay or paper with the assignment marking tool.

Comment

Assignment Marker - AI-Powered Grading Assistant for Teachers

Inspiration

After watching my partner, a French and English teacher, dread marking long-form assignments for years, I knew something had to change. She would spend over 10 minutes per essay, multiplied by 30+ students per class makes 5+ hours of marking per class. With 3 classes and 10 or more assignments per semester, the hours stack up. When I interviewed dozens of other teachers, I discovered this wasn't just her problem, it was a universal pain point affecting educators worldwide.

This project has been stewing in my mind for over 4 years, but thanks to Bolt.new, I was finally able to turn this vision into reality in just 2 weeks instead of months of traditional coding.

What it does

Assignment Marker is a privacy-first AI grading assistant that transforms how teachers approach long-form assessment. Teachers simply upload their rubric and assignment instructions, then submit student work (typed text, handwritten papers via OCR, or PDFs). The AI analyzes everything contextually and provides:

  • Accurate grades aligned with the teacher's specific rubric
  • Structured feedback organized by assessment criteria
  • Time savings of 70%+ (10+ minutes reduced to ~3 minutes per assignment)
  • Consistency in grading standards across all students

The tool specializes in language arts assignments -essays, creative writing, long-answer responses - where subjective grading traditionally takes the most time.

How we built it

Frontend Architecture:

  • React 18 with TypeScript for type safety
  • Tailwind CSS for responsive, teacher-friendly UI
  • Vite for lightning-fast development builds
  • Custom React hooks for file upload and state management

AI Pipeline:

  • OpenAI GPT-4o for vision-based OCR (reading terrible handwriting!)
  • Google Cloud Vision API for faster OCR on easier handwriting
  • GPT-4o-mini for content analysis and grading
  • Multi-step AI workflow: content analysis → assignment extraction → grading
  • Custom prompt engineering to ensure fair, encouraging feedback (initial prompts were way too harsh!)

Backend & Security:

  • Netlify Functions for serverless API proxy
  • Secure environment variable handling to prevent API key leaks
  • Local storage for privacy - no student data is ever saved on our servers and is not used for training.

Key Technical Achievements:

  • End-to-end OCR pipeline that actually works with student handwriting
  • Contextual AI that understands assignment requirements before grading
  • Graceful error handling throughout the complex AI workflow
  • Teacher-centric UX designed around real marking workflows

Bolt.new was instrumental in accelerating development - it generated the entire application framework, handled data flow between components and Netlify Functions, debugged complex issues, and responded perfectly to UI feedback. What would have taken months became weeks.

Challenges we ran into

  1. Prompt Engineering: Initial AI responses overly harsh. Spent most of my time refining prompts to be fair, encouraging, and aligned with modern pedagogical standards.

  2. Error Handling Complexity: With multiple AI API calls, OCR processing, and file uploads, creating graceful fallbacks for every failure point was challenging but crucial for teacher trust.

  3. Secure Deployment: Ensuring API keys stay secure in Netlify while maintaining functionality across different environments required careful architecture planning.

Accomplishments that we're proud of

  • Real Impact: Tested with teachers and achieved 70%+ time savings (10+ minutes → 3 minutes per assignment)
  • Accuracy: Hitting 80%+ matching with benchmark assignments on evaluation tests
  • Privacy-First: Built a powerful AI tool that never stores student data on servers
  • Teacher-Centric Design: Unlike admin-focused tools, this is built specifically for how teachers actually work
  • Technical Achievement: Created a seamless OCR-to-grading pipeline that handles messy student handwriting

What we learned

  • User Research is Everything: Teachers have been ignored by edtech for too long. Building FOR teachers, not just for administrators or students, completely changes the product.
  • AI Prompt Engineering is an Art: Getting AI to grade like a caring, experienced teacher (not a harsh critic) requires deep understanding of both pedagogy and prompt psychology.
  • Privacy Enables Adoption: Teachers are much more willing to try tools that don't require uploading sensitive student data.
  • Bolt.new is a Game-Changer: This project proved that AI-assisted development can compress years of work into weeks without sacrificing quality.

What's next for Assignment Marker

Immediate Roadmap:

  1. Multi-Model Grading: Implementing Grok, Qwen, and Llama models with a consensus algorithm for even more accurate grading
  2. Batch Processing: Multi-student marking to upload an entire class worth of assignments for simultaneous grading
  3. Better Data Management: Improved assignment library and export capabilities based on teacher feedback

Long-term Vision: Save millions of hours for teachers worldwide. Back-of-napkin math: Just 5,000 active users could save over 1 million hours annually. With 400,000+ teachers in Canada alone (and 10x that in the US), the potential impact is enormous.

Business Model: While built with love for educators, this represents a significant business opportunity. The goal is to start free for teachers, then explore sustainable monetization that serves the education community.

Teachers are incredible humans doing impossible jobs with insufficient support. In an era of budget cuts and growing class sizes, tools like Assignment Marker aren't just nice-to-have—they're essential for sustainable education.

Built solo in 17 days with Bolt.new for the Hackathon.dev AI Vibecoding challenge.

Built With

Share this project:

Updates

posted an update —

Update The application now has a new UI for submitting assignments and receiving the feedback! It now uses a chat interface after the user creates the assignment. This means that the user can see the full context of their previously marked assignments and easily export all of the data from their marked assessments.

Log in or sign up for Devpost to join the conversation.