Most conversations about AI in education still centre on chatbots: you type a question, you get an answer, and then a member of staff does all the actual work of turning that answer into a homework, a report or an intervention. Useful, but it leaves the hard part exactly where it was.
Raphael is different. Raphael is the institution's AI teaching agent, and it is built to run the workflows your teachers and leaders already run, from start to a reviewable finish. Staff ask it for what they need, and it comes back not with a wall of text but with structured, finished work that a person checks, edits and approves.
What Raphael actually is
Raphael sits on top of ilmino, the AI learning management system used across your school, college, university or trust. Where ilmino provides the underlying capabilities (AI marking of typed and handwritten work, a question and test builder, a curriculum workspace, department and trust analytics, and integrations with your existing systems), Raphael is the layer your staff talk to.
You do not need to learn where every feature lives or click through a dozen screens. You describe the outcome you want, in plain language, and Raphael assembles it for you.
Think of it less as a search box and more as a highly capable teaching assistant who knows your curriculum, your classes and your data, and who can draft real work for you to sign off.
How staff use it: chat or voice
Getting work from Raphael looks like a normal conversation.
- By chat: a head of department types a request in everyday language, for example "build me a term of homework for Year 11 Physics."
- By voice: a busy classroom teacher speaks the same instruction between lessons, hands free.
There is no special syntax to memorise and no prompt engineering to learn. If you can describe what you want to a colleague, you can ask Raphael.
What it returns: action cards, not walls of text
This is the part that changes the day to day. Raphael does not answer with paragraphs you then have to act on. It returns structured action cards: discrete, reviewable units of finished work, each with a clear proposed action.
An action card might say:
- Schedule this homework to 11X/Ph on Monday.
- Publish this progress report to the department.
- Open this intervention list for the five students flagged below.
Each card is editable. A teacher can adjust a question, change a date, swap a class, or reword a comment before anything happens. Nothing is committed simply because Raphael proposed it. The card is a draft with a button, and the person decides.
The specialist agents behind one instruction
Here is the idea that makes Raphael feel different, explained without the jargon.
When you give Raphael a single instruction, it does not try to do everything itself in one pass. It breaks the request down and hands the pieces to a set of specialist agents, each good at one job, then brings the results back together as finished cards. Behind one plain-language request, Raphael may coordinate agents for:
- Question generation: producing questions matched to the topic, tier and difficulty you need.
- Marking: applying mark schemes to typed and handwritten work.
- Feedback writing: drafting comments in a consistent, constructive voice.
- Lesson and slide creation: building teaching materials aligned to the same content.
- Analytics: reading class and cohort performance to inform what comes next.
- Intervention planning: identifying who needs support and proposing a targeted list.
- Scheduling: placing work on the right dates for the right groups.
- Integrations: connecting to your school information system, LMS and other tools so nothing has to be re-entered.
You never have to think about which agent does what. You ask once; Raphael orchestrates the rest. The analogy that tends to land with leadership teams is a well run office: you give one clear instruction to a capable manager, who quietly delegates to the right specialists and returns to you with the completed work ready for sign-off.
Long-range planning in one go
Because Raphael completes whole workflows rather than single answers, it handles scale that would be tedious by hand. You can ask it to plan across time, not just for the next lesson.
A department can request a full term of homework, or a sequence of assessments across a unit, in a single instruction. Raphael drafts the whole run, spaced sensibly and mapped to your curriculum, and returns it as a set of cards you can review together. Planning that used to take an afternoon becomes a review exercise measured in minutes.
A worked example
Consider a head of department who says, by chat or voice:
"Set up a term of Year 11 Physics homework, one task a week, covering electricity and forces, higher tier, and flag anyone likely to struggle so I can plan support."
Behind that one instruction, Raphael coordinates its specialist agents and returns a stack of action cards, for example:
- Card 1 (Homework plan): twelve weekly Year 11 Physics tasks on electricity and forces, higher tier, each with questions, mark schemes and suggested dates. Actions: edit, reschedule, approve.
- Card 2 (Slides): a matching set of lesson slides for the opening electricity topics. Actions: edit, publish to shared area.
- Card 3 (Marking setup): auto-marking configured for each task, including handwritten responses. Actions: review settings, approve.
- Card 4 (Intervention list): eight students flagged from recent performance as likely to need support, with the topics they are weakest on. Actions: edit list, open intervention.
- Card 5 (Analytics view): a progress dashboard that will track the cohort across the term. Actions: share with department.
The head of department reads through, tweaks a couple of questions, moves one deadline around a mock exam, removes a student who has since improved, and approves. Only then does anything reach a class.
Why an agent that finishes work is different from a chatbot
A chatbot answers. An agent completes.
- A chatbot can describe a good homework plan. Raphael builds one, schedules it, sets up the marking and drafts the interventions, then hands it to you to approve.
- A chatbot leaves the assembly, the clicking and the data entry to your staff. Raphael does that part and asks a person to review the result.
The difference is where the effort lands. With a chatbot, the human does the work after the answer. With Raphael, the human reviews the work instead of producing it. For teachers under time pressure, that is the difference between a clever tool and a genuine reduction in workload.
Control and governance: nothing reaches a class without approval
Because Raphael acts on real classes and real data, the control model is deliberately strict.
- Human approval is required. Every action card is a proposal. No homework is set, no report is published and no intervention is opened until a member of staff approves it. AI assists; teachers and admins review, edit and approve every action.
- Roles and permissions are respected. Raphael works within your existing structure. A teacher sees and acts on their classes; a head of department on their subject; a senior leader or trust lead on their wider view. Raphael never lets someone act beyond their remit.
- Everything is reviewable. Because work arrives as discrete cards rather than opaque output, leaders can see exactly what was proposed, what was changed and what was approved.
This is what makes an agent safe to put in front of a real institution: capability with a person firmly in the loop.
Getting started
Introducing Raphael does not mean changing how your institution teaches. It sits on the ilmino platform your staff already use, respects the roles and permissions you already have, and starts by drafting work your team was going to do anyway.
A sensible first step is to pick one department and one workflow, for example a term of homework or a round of assessments, and let staff experience the shift from producing work to reviewing it. From there, most institutions widen use across subjects and phases at their own pace.
If you lead a school, college, university or trust and want to see Raphael handle your real workflows, book a demo and we will walk your team through it with your own use cases in mind.
