FP Smart Tutor Offline

• 

  OffLine VET

• 

  Ohm Law

Inspiration

The project was inspired by the digital divide that still affects many schools and training centers: Some classrooms do not have stable internet. Teachers are concerned about privacy when using cloud-based AI tools. Students in Vocational Training often need personalized support but lack individual tutors. We wanted to create a tool that ensures AI for everyone, everywhere.

What it does

EduGPT-Local is an offline AI tutor that runs entirely on local hardware (laptops or Raspberry Pi) using gpt-oss models. It is designed for Vocational Education and Training (VET/FP), and it can: Answer student questions in real time, even without internet access. Generate exercises and quizzes, then check the answers. Provide didactic explanations of technical concepts (e.g., electromedicine, telecommunications, commerce). Work on low-resource devices, making AI accessible to any classroom.

How we built it

Dataset preparation We created an educational dataset with open materials and our own content. Example of a simple physics law included: V=I⋅R V=I⋅R (Ohm’s Law, explained in an accessible way for students). Fine-tuning We applied LoRA adapters to optimize gpt-oss models for domain-specific tasks, focusing on concise, didactic explanations. Local deployment We used llama.cpp to run models offline on both laptops and Raspberry Pi 5, making the tool accessible for low-resource environments. User Interface A lightweight web interface (HTML/JS frontend + Python backend) that works directly in the browser at localhost.

Challenges we ran into

Model size vs. hardware limits: balancing performance so that the AI can run on devices with only 8GB RAM. Data curation: ensuring that the educational dataset is both accurate and adapted to vocational students. Multilingual support: enabling responses in English and Spanish. Time constraints: building a working demo in just a few weeks.

Accomplishments that we're proud of

🚀 Running gpt-oss fully offline on modest hardware (including Raspberry Pi 5), proving that powerful AI does not need cloud services. 📚 Creating a domain-specific dataset for Vocational Training (Electromedicine, Telecommunications, Commerce) and successfully fine-tuning with LoRA. 🖥️ Building a lightweight user interface that teachers and students can use without technical knowledge. 🌍 Supporting inclusivity and accessibility, giving AI tools to classrooms without stable internet access. 🔒 Guaranteeing privacy by ensuring no data ever leaves the local device. 🏫 Testing with real students in vocational education, who provided positive feedback about clarity, speed, and usefulness. 🤝 Collaborative team effort: developers, teachers, and students worked together across disciplines to make the project real.

What we learned

How to deploy gpt-oss models locally using lightweight frameworks like llama.cpp. How to fine-tune models with LoRA (Low-Rank Adaptation) to specialize in domains like Electromedicine, Telecommunications, and Commerce. The importance of designing simple user interfaces that teachers and students can use without technical barriers. How to balance accuracy vs. efficiency when running AI on limited hardware such as Raspberry Pi.

What's next for FP Smart Tutor Offline

🔧 Expand fine-tuned datasets to cover more vocational fields, including healthcare, industrial automation, and retail services. 📱 Mobile version: create a lightweight Android app so students can carry the offline tutor in their pocket. 🗣️ Voice interaction: integrate speech-to-text and text-to-speech for accessibility and inclusive learning. 🌐 Multilingual support: extend beyond English and Spanish to other European and global languages. 🤝 Open collaboration: publish our datasets and fine-tuning recipes so other educators can adapt the tool to their own contexts. 🏫 Pilot programs: deploy EduGPT-Local in real vocational classrooms and collect structured feedback from teachers and students. ⚡ Optimization: improve inference speed to run smoothly on devices with as little as 4 GB RAM. 👉 Our vision: make offline AI tutors a standard tool in vocational education worldwide.

Built With

• choicescript
• html