iFundoPlus

• 

  architecture diagram

🎯 Inspiration

As a full-time teacher and solo developer, I’ve seen how education gaps deepen where connectivity is poor. Many rural learners can’t access cloud-based AI tools due to limited internet, device costs, or privacy concerns. I built IfundoPlus to bring AI-powered learning to places where it typically can’t reach — using low-cost hardware, offline-first design, and governance-focused safety controls.

🚀 What it does

IfundoPlus is a voice-controlled educational assistant that works both online and offline. It answers curriculum-aligned questions, runs flashcard-based recall training, and supports past paper revision — all through natural voice interaction. A built-in GovLayer filters unsafe content in real time, ensuring classroom safety and compliance. The app switches automatically between online (Amazon Bedrock, Polly) and offline (local embeddings, pyttsx3) modes.

🛠️ How I built it

The backend is built in Python using Flask, running on either Raspberry Pi or EC2. The assistant uses:

• Amazon Bedrock (Titan Text G1 – Express) for online generation
  
• Titan Embeddings G1 for offline RAG-style Q&A
  
• Vosk for offline STT and Amazon Polly/pyttsx3 for voice responses
  
• GovLayer for query filtering and policy enforcement
  
• S3 for flashcard/past paper storage
  
• Crontab to autostart on boot
  

All logic is modular, with fallback layers that adapt based on internet availability.

⚠️ Challenges I ran into

• Implementing real-time governance filtering without cloud moderation tools
  
• Tuning the voice interaction to feel natural while keeping latency low offline
  
• Keeping RAM and storage usage minimal for Raspberry Pi compatibility
  
• Testing behavior across changing network conditions with no external logging

🏆 Accomplishments that I'm proud of

• Fully voice-operated, offline-compatible RAG system using Titan embeddings
  
• Custom-built GovLayer that flags or blocks unsafe queries before generation
  
• End-to-end solution running under 400MB RAM on a Pi
  
• A real tool tested with students, showing improved recall and engagement in real classrooms

📚 What I learned

This project taught me how to integrate RAG, multi-agent fallback, and governance enforcement into a cohesive system. I deepened my experience with AWS Bedrock, real-time filtering logic, and privacy-first design. Most of all, I learned how to build tech that works for students — not just users.

🔮 What's next for iFundoPlus

Next steps include:

• Adding support for multilingual Q&A
  
• Expanding GovLayer to use context-aware NLP filtering
  
• Building a teacher dashboard for usage analytics and content control
  
• Exploring deployment at scale in rural schools across Africa
  
• Publishing a research paper on the impact of offline-first AI in education

Built With

• amazon-bedrock-(titan-text-g1-express-&-titan-embeddings-g1)
• amazon-ec2
• amazon-polly
• amazon-web-services
• and
• crontab
• database
• embeddings
• embeddings.json
• external
• flask
• govlayer-(custom-ai-governance-filter)
• in
• local
• no
• python
• pyttsx3-(offline-text-to-speech)
• raspberry-pi-4b
• stored
• titan
• vector
• vosk-(offline-speech-recognition)