Course-material tutors, admissions and administration copilots, lecture transcription and early-warning analytics for schools, universities, training providers and ed-tech companies, built around student data protection and welfare.
The best education AI gives staff time back and gives learners a patient explainer, without ever becoming the authority on a student's welfare.
Course-material tutors
A per-course assistant that explains concepts from the lecturer's own slides, readings and past papers, cites where the answer came from and asks the learner questions back. Touches the LMS and the course content repository. Behaviour is configured per course and assessment.
Admissions and student-services copilots
Answer entry requirements, deadlines, fees, accommodation and timetable questions from approved sources, and hand complex cases to staff with context. Touches the CRM, the student information system and the website. Built on our customer support AI pattern.
Lecture transcription and accessibility
Captions, transcripts and summaries for lectures and seminars, searchable by topic, supporting students with accessibility needs and anyone revising. Touches lecture capture and the LMS. See speech-to-text.
Feedback drafting for markers
Draft formative feedback against a rubric for the marker to edit and own, with consistency checks across a cohort. Touches the assessment platform. The grade stays with the human; the time saved goes back to teaching.
Policy and procedure assistant for staff
Grounded answers on attendance, assessment regulations, HR policy and safeguarding procedure, with the source paragraph and version. Touches the intranet and document store. See internal copilot.
Early-warning analytics
Attendance, submission and engagement signals combined into a prompt for a tutor to check in, never an automatic consequence. Touches the student information system and the LMS. Transparent features, reviewed by the people who act on them.
Compliance and data
Safeguarding, student data and fairness
What we design for in education
Students are often children or young adults, so their data gets the highest protection under GDPR and UK GDPR, and age-appropriate design applies to anything they interact with. Safeguarding duties mean an AI assistant must recognise welfare signals and route them to a person immediately, not attempt to help. Automated decisions about a learner's progress need human review, and any early-warning model must be explainable to the staff who act on it. Accessibility standards apply to the interface. This is engineering guidance, not legal advice; your data protection officer and safeguarding lead own the policy.
We keep student data out of model training, deploy privately where required, log every student-facing interaction for authorised review and test safeguarding filters with adversarial prompts. The AI governance service produces the documentation governors and auditors ask for.
Reference architecture
Flagship workflow: a course tutor with guardrails and a safeguarding route
The learner asks from the LMS or the institution's app. The assistant first checks enrolment against the student information system, so it only discusses courses the student is on. Every question passes a scope and safeguarding filter on the way in and every answer on the way out.
Answers come from an index of the course's own materials, cited by slide or page. Usage, confusion hotspots and flagged questions appear on the teacher's dashboard, which is where most of the pedagogical value appears: the questions students ask show what the teaching missed.
If a message suggests a welfare concern, the assistant does not counsel. It responds with a short, safe message and alerts the designated safeguarding lead with the conversation. The pattern is a specialised form of our enterprise RAG work with the guardrail design moved to the front.
The assistant explains course content; it never becomes the first responder for a student's welfare.
Track record
What we have built that applies
Learning and engagement apps
Native mobile apps with content delivery, progress tracking, notifications and offline access, drawn from two decades of app delivery across sectors.
How do you stop a student assistant giving harmful or off-topic answers?
Three layers: the assistant only answers from the course materials it is given, every input and output passes a safeguarding and scope filter, and anything that suggests a welfare concern is routed to the designated safeguarding lead rather than answered. Teachers see usage and flags on a dashboard. We test the filters with adversarial prompts before launch; see AI red teaming.
Will students use it to cheat?
A tutor that explains and asks questions back is different from one that writes the essay. We configure the assistant's behaviour per course: hints and worked examples where learning is the goal, no full solutions on assessed tasks, and logs that academic staff can review. Policy is yours; we make the tool follow it.
Can it run on-premise or within our existing cloud tenancy?
Yes. Many institutions prefer models that run inside their own cloud tenancy or on-premise, especially for children's data. We deploy open-weight models where needed and hosted models under agreement where allowed. See on-prem versus cloud LLMs.
Where should an institution start?
Administrative load is the safest first win: admissions enquiries, timetable and policy questions, lecture transcription for accessibility. Student-facing tutoring follows once the data and safeguarding controls are proven on lower-risk use.
Ready to make AI work in production?
Tell us about your systems, your data and your constraints. We reply within one business day with a straight answer and a suggested next step.