MaestroMaestro Overview — contents
The Concept
The AI Companion
The reflective learning partner that lives on the page throughout the course. It guides, explains adaptive recommendations, and checks understanding without handing over answers — and, crucially, it is not an open assistant. Every interaction is structurally bounded.
The Companion is on the page the whole time and knows where the learner is. It gives honest feedback and prompts reflection instead of handing over answers, reads the node metadata, and is expected to feel real-time. Its core philosophy: increase learner awareness, not learner dependency. It is not a free-form tutor — every interaction sits inside an authored Learning Anchor with a defined purpose and explicit forbidden moves.
What it is
The Companion is a reflective learning partner, not a tutor and not an answer engine. It is present on the page throughout, aware of the learner's current node, and it reads the node metadata so it can talk to the learner, guide them, and check what they know. It is expected to feel real-time (likely over real-time APIs — a build decision noted as open). And it picks from teaching approaches people have written; it does not invent teaching on the fly.
| The Companion intentionally avoids | The Companion does |
|---|---|
| Spoon-feeding AI tutors · answer-generation systems · passive AI dependency | Encourages reflection & metacognition · explains adaptive recommendations · reduces isolation in self-paced learning · helps learners navigate challenge intentionally |
It is also a journey advisor: it helps the learner understand where they are — CLO progress, node readiness, unresolved misconceptions, the recommended next step, available challenges, assessment readiness, and how progress connects to badges, Mastery Credits, or contribution opportunities. It keeps learner agency visible rather than deciding for them.
Instead of "You failed. Redirecting to remediation," the Companion says: "You are explaining what the framework includes, but critical evaluation requires judging its value using criteria. Let's practice turning description into critique." Transparent feedback that encourages reflection and preserves agency.
Learning Anchors
A Learning Anchor is authored, bounded in-journey guidance. Anchors are either fully authored (fixed text) or runtime handoff templates that the live Companion composes against at runtime, within the anchor's boundary. Each anchor has a defined purpose — orientation, transition, remediation introduction, enrichment invitation, reflection prompt, and so on — and the Companion can only operate within that purpose. Anchors travel in the Published Course Package so the runtime knows exactly what the Companion is allowed to say at each point.
The bounded architecture
Because the Companion is central and must never become a back door to answers, it is bounded by four structural constraints at once — anchor scope, forbidden moves, content-spec grounding, and runtime handoff bounds — plus a behavioural steering-attempt layer. These are covered in full, from the integrity angle, in Academic Integrity. The short version: it has no free-form mode, it cannot see Evidence Check answers or rubric criteria, and it composes only against the evidence record, the learner-model state, and the anchor's template.
Two rendering modes: authored-anchor rendering for fixed anchors, and runtime-dynamic composition for handoff templates — composed against the evidence record, learner-model state, and message template. Its context is the approved Level 2 content spec for the current node; it must not receive Evidence Check answer keys, rubric grading criteria, or milestone submissions. Runtime governance metadata pins the prompt/judge versions the course was authored against, so the Companion behaves deterministically against that version, not the newest prompt.
The adaptive flows
The Companion's behaviour is built as clear flows, most-obvious-cases first, widened by real usage — with the intent that the learner always has a companion alongside them, with no gaps. Two flows are the starting point:
| Flow | What happens |
|---|---|
| After an assessment | The Companion explains in plain terms what the learner got wrong (e.g. "you seem to understand the terms but not yet how to apply them") and offers a short list of personalised next steps: redo the node · watch a worked example · review a prerequisite · take a guided walkthrough · retry. |
| In-content help | The learner selects a passage that isn't clear, and the Companion explains it right there, in place. |
Behaviour shifts by band
The Companion adapts its tone and framing to the diagnostic band and the node type rather than treating every moment the same way: supportive after a misconception is surfaced, celebratory but stretching after secure mastery, orienting at node entry, and connecting to the milestone as readiness approaches. It reads the learner-model state to decide how to show up — this is itself one of the eight adaptive mechanisms.
Dialogic verification
At the highest-stakes moments, the Companion runs the dialogic probe — a Companion-led reflection that confirms mastery is real through conversation, not accusation. It is the final layer of the integrity model: where a diagnosis needs confirming, a short reflective exchange separates genuine understanding from a fluent surface answer, and its outcome feeds the diagnostic layer. This is the sense in which the Companion is not just support but part of how Maestro believes a learner has mastered something.
The Companion is a runtime component of Maestro Journey — it exists because a specific learner is interacting. It reads the learner model and the current node's approved content spec, composes within authored anchors, and writes interaction patterns back as learning signals. What it can and cannot see is deliberately narrow, and that narrowness is the point.