Module 18: Who uses this: sectors and careers · Lesson 18.4 · about 20 min
Teachers, students and offline classrooms
Local AI in schools and colleges, from a teacher's laptop to a classroom server with no internet. Uses, child-safety and privacy points, and a setup that works where connections are poor.
Module 18 · about 20 minutes · read first: 9.5 and 9.6. You’ll be able to plan local AI for a teacher, a class or a school.
1. Why this sector
Teachers want help with preparation and marking; students want a patient tutor. Many schools have weak or costly internet, and student data needs care. A local model on a laptop or a small classroom server works offline and keeps student work in school.
2. Top uses
- Teachers: worksheets and question banks at several levels, rubrics, feedback drafts on student work, simpler explanations for different learners, parent letters.
- Students: step-by-step help on homework (with the teacher’s rules), practice quizzes, language practice, explaining a paragraph in simpler words.
- Offline classrooms: one server in the school, reached over Wi-Fi from cheap laptops or tablets, so a whole class can use AI with no internet at all.
3. Care points
- Children’s data: extra legal protection in all four countries (DPDP parental consent in India, COPPA and FERPA in the USA, UK GDPR’s children’s code, provincial laws in Canada). Local keeps it in school; still use accounts and short retention.
- Content: use instruction-tuned models from major publishers, a clear system prompt for age and tone (Lesson 10.2), and teacher review.
- Honesty: set rules for when AI help is allowed and how to say it was used.
4. A sensible setup
Teacher: a 16 GB laptop with Gemma 4 E4B or 12B (Module 9). Classroom: a 32 to 64 GB machine running a 12B model with batching for 30 students (Lessons 15.1 and 15.2) and Open WebUI with class accounts, on the school Wi-Fi only. Batching means the server writes several answers at the same time instead of one after another.
A class asks in bursts, like the queue at a school canteen in the first five minutes of lunch:
The activity starts. 30 students open the classroom chat. Half of them, 15, ask a question in the same minute.
One at a time. Each answer is about 300 tokens and the server writes 40 tokens per second: 7.5 s per answer. Without batching, answers come one after another.
The last in line. The 15th student waits for 15 answers: 15 × 7.5 = 112.5 s, nearly two minutes. In a 40-minute lesson, that feels broken.
Turn on batching. With 8 parallel slots, the server writes 8 answers at once, each a little slower (say 25 tokens per second, 12 s per answer). Two rounds serve all 15 in about 24 s.
Maths and principle
Principle: a class asks in bursts, so plan for the start of the activity.
Equation: peak load ≈ students × share asking in the same minute.
Analogy: the queue at a school canteen in the first five minutes of lunch.
Try it: 35 students, half asking within the same minute, 300-token answers at 40 tokens per second without batching? (About 17 answers × 7.5 s ≈ 2 minutes for the last student: turn on batching.)
- S ÷ 2
- students asking at the same moment (here, half the class)
- T
- tokens in one answer
- v
- the server's writing speed in tokens per second, one answer at a time
- tlast
- how long the last student in the queue waits, without batching
- Physical meaning
- A server that feels instant for one teacher can feel frozen for a class, because everyone arrives at once. Size the server for the rush at the start of an activity, not for the average.
“The AI tutor explained it clearly, so the answer must be right.” A clear explanation can still be wrong (Lesson 2.7), and small models make more mistakes. Teach students to treat AI answers as a starting point and check them against the textbook or the teacher. Learning to check is part of using AI well.
5. Worked scenario
Made-up school. A government higher secondary school in a hill district with slow internet installs a 64 GB mini PC in the computer lab, running a 12B model with 8 parallel slots. Students use it for practice questions in Physics and English in lab periods. Teachers prepare worksheets on it at school. No student work leaves the lab.
6. Practice
- Write a system prompt line that sets the age and tone for your class.
- What should a student do with an AI answer before using it?
- Why do accounts matter even in an offline classroom?
Answers
- For example: “You are a patient tutor for 14-year-olds. Explain step by step, ask a question back, never give the final answer to homework.”
- Check it against the textbook or ask the teacher.
- Each student’s chats stay private, and the teacher can see who needs help.
New words: classroom server, age-appropriate, parental consent.
- A teacher’s 16 GB laptop or one classroom server on the school Wi-Fi works with no internet at all.
- Children’s data has extra legal protection in all four countries: use accounts and short retention even offline.
- Set age and tone in the system prompt, and rules for when AI help is allowed.
- Classes ask in bursts: without batching the last student waits (students ÷ 2) × (tokens ÷ speed); turn batching on.
- AI answers are a starting point; students check them against the textbook or teacher.
Check yourself
Sources
- UNESCO, "Guidance for generative AI in education and research", 2023 (official page), https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research
- UK Department for Education, "Generative AI in education" (official page), https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education
- India: Digital Personal Data Protection Act 2023, verifiable parental consent for children's data (MeitY), https://www.meity.gov.in/data-protection-framework
- USA: FERPA and COPPA (Department of Education and FTC, official pages), https://studentprivacy.ed.gov and https://www.ftc.gov
Your ticks stay in this browser. Save them as a file on the course home.
Updated 2026-11-22. Prices and model names change fast: check the official source before you buy.