Offline AI Tutor Hotspot on a Raspberry Pi for Schools Without Internet
Many schools have a few phones or tablets but no reliable internet. Build a Raspberry Pi box that runs a small open language model and a curriculum-grounded retrieval system completely offline, creates its own Wi-Fi network, and lets students ask questions about their textbook in Urdu or English from any browser. Teachers control what content is loaded and see which topics students ask about most.
Intermediate
A small team, or one strong student willing to learn something new.2 semesters · 2 to 3 students
7 to ship
3 optional extrasSuggested stack
What you should ship
- A Raspberry Pi image that boots into a Wi-Fi hotspot and serves a local web app, with no internet needed after setup
- Retrieval over loaded textbook content (openly licensed or teacher-provided) with answers that quote the page they came from
- A small quantised open model chosen by benchmarking three candidates on the Pi for speed, memory and answer quality
- A queue and fair-share system so 20 students on one Pi each get an answer in reasonable time
- A teacher dashboard: load content, view common questions, flag bad answers
- Safety rules: answers only from loaded content, a refusal when nothing relevant is found, and no data leaving the device
- A field-style test with at least 15 students (with school and parent permission) measuring answer time, accuracy and usefulness
If you have time left
- Voice questions with an offline speech model
- Content sync from a USB stick or when a phone with data is briefly connected
- Running on solar power with a battery and measuring daily energy use
The problem: AI help that needs a data connection
AI tutors and chatbots assume a fast, cheap internet connection and a personal device. Many government and low-fee private schools across Pakistan, Bangladesh and Africa have neither: a few shared phones, patchy mobile data and frequent power cuts. Students who most need help are the least able to use cloud AI.
Small open models can now run on a single-board computer. A Raspberry Pi 5 is slow compared with a cloud GPU, but it is cheap, quiet and can run on a battery.
What already exists and the gap
- Kolibri by Learning Equality is an offline-first learning platform used in low-connectivity schools worldwide.
- Internet-in-a-Box and RACHEL serve Wikipedia, Khan Academy videos and other content over a local Wi-Fi network.
- Ollama and llama.cpp make running small models on modest hardware straightforward.
Offline content servers do not answer questions, and AI tutors need the cloud. The gap is a box that does both: grounded question answering over the school's own curriculum, offline, for a whole classroom at once.
Inspiration
Offline content servers have proven that a cheap box in a classroom can change what students can access. This project tests whether adding a small, grounded language model is practical on the same kind of hardware.
What you build
- A bootable Pi image with hotspot, web app and model.
- Grounded retrieval that always shows its source.
- A classroom-scale queue and a teacher dashboard.
The technically interesting part
- Choosing a model for 8 GB of RAM. Benchmark three small models for tokens per second, memory and quality on curriculum questions. The best model on paper may be too slow for a class.
- Many users, one small computer. Short answers, streaming and a fair queue matter more than raw speed.
- Grounding. If nothing relevant is in the loaded content, the model must say so instead of inventing an answer.
How you will evaluate it
- Answer accuracy on 100 curriculum questions checked by a teacher.
- Median and worst wait time with 5, 10 and 20 simultaneous users.
- Power use per hour and per answer.
- Feedback from students and a teacher in a supervised session.
Path to a product
NGOs, education departments and school chains run device programmes and need offline tools. A preconfigured box with content packs and support, or an open source image with paid setup, are realistic models.
Scope warning
Use openly licensed or teacher-provided content and respect copyright. Get permission from the school and parents before any session with children and collect no personal data. Do not fine-tune models; choose and benchmark existing ones.
Learn what you need on Sefism
- Embedded and IoT roadmap: Raspberry Pi setup, networking and power.
- AI engineering roadmap: retrieval, grounding and choosing a model.
- Raspberry Pi documentation: hotspot setup and performance tuning.
- Ollama: running small models locally.
- Choosing LLMs cheat sheet: how to compare small models on speed and quality.
- Arduino, ESP32, STM32 or Raspberry Pi: which board to start with: why the Pi is the right board for this job.
- How do I choose which LLM to use for my project?: how to pick a small model that fits the task.
Get help with this idea
The risk is a box that works for one user and stalls with a class. Tauseef can help you choose models, plan the load tests and arrange a responsible school trial. If you are new here, create a free account first.
Ideas and guidance, not finished projects
These are project ideas and scoping guidance, published free for students to use as a starting point. I do not build, write, or sell final-year projects, and I do not complete coursework for anyone. Take an idea, make it yours, and build it.
Not sure this one fits you?
I help students pick an idea that matches their skills and their deadline, then scope it down until it is finishable. Guidance only, never done for you.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.