graspy

Education that reaches you, wherever you are

Comment

Inspiration

244 million children worldwide are out of school. In Nigeria alone, over 10 million are missing from the classroom. War, poverty, and distance have shut the school gates. Traditional education can’t reach them: but technology can.

Existing digital solutions are Western-centric, English-only, require stable internet, and assume continuous schooling—failing those who need them most.

With graspy, we ensure every child, regardless of circumstance, has access to a personal AI tutor.

What It Does

graspy is an AI-powered tutor that creates personalized learning paths for students wherever they are, in their own language. It's always available to adapt to each student's cognitive level, cultural context, and educational needs—continuously refining their learning journey based on performance.

How It Works

  1. Onboarding: Student provides country, age/grade, and preferred language
  2. Subject Generation: AI agents create a curated list of approved subjects and courses relevant to their education system
  3. Subject Selection: Student chooses what they want to learn
  4. Curriculum Design: Multi-agent AI generates a personalized learning path adapted to their country's curriculum (Nigerian WAEC, Syrian Ministry of Education, etc.), language, and grade level
  5. Contextual Lessons: On-demand lesson generation with offline storage. Lessons use local currency, food, and cultural contexts—not generic Western examples
  6. Real-Time Mentoring: Multilingual AI chat (10+ languages) for students to ask questions about their current lesson

How We Built It

Frontend: Next.js React with i18n support for English and Arabic

Backend: Python FastAPI exposing agent capabilities via REST APIs

LLM: AWS Bedrock for LLM inference; AWS Nova Lite as our model

Agent Architecture (orchestrated via AWS Strand):

  • Subject Generator Agent: Generates available subjects based on country and education system
  • Curriculum Designer Agent: Creates personalized learning paths for each subject
  • Lesson Orchestrator Agent: Coordinates two sub-agents:
    • Lesson Content Generator: Produces slide-based lesson content for each topic
    • Assessment Generator: Creates contextual assessments for each lesson
  • Mentor Chat Agent: Answers student questions using the current lesson as context

Challenges We Ran Into

  1. Managing token limits when generating long-form lesson content
  2. Reducing latency when orchestrating multiple lesson generation requests

Accomplishments We're Proud Of

  1. Personalized on-demand content generation across multiple languages supported by Amazon Nova.
  2. Rendering mathematical formulas using LaTeX directly within LLM outputs

What We Learned

  1. Using agents as tools within orchestrating agents for complex workflows
  2. Leveraging AWS Bedrock effectively for multi-agent systems
  3. Building scalable agent and tools with AWS Strand

What's Next for Graspy

Short term (1-3 months):

  • Launch a stable MVP with richer, more usable content
  • Implement content caching to reduce redundant LLM calls
  • Begin pilot sessions with out-of-school communities in Nigeria (through local NGO partnerships)
  • Test and validate additional languages

Long term (3-year goals):

  • Access: Reach 1M+ students across crisis zones
  • Learning: Achieve 1-2 grade level advancement in 6 months
  • Engagement: Maintain 60%+ topic completion rates and 50%+ of users in 3+ sessions/week

Built With

Share this project:

Updates