AI infrastructure for education

Personalised learning only works if the system remembers the learner, and most education tools don't. A tutoring feature that starts every session from zero can't tell what a student has mastered, where they're stuck, or how they learn best, so the personalisation that makes tutoring effective never gets going. If you're building learning platforms, the infrastructure you need underneath is per-student memory that's isolated and persistent, a way to turn lectures and course material into something searchable, and search that finds by meaning.

This page is for teams building education and learning products. Exabase gives you per-student isolation and memory through Bases and Memory, lecture and material transcription through Extract, and queryable course libraries through Deep Search. Every learner gets their own isolated picture of what they know and how they learn, at the scale of a whole platform. For the full tutoring build, the learning and tutoring copilots use case goes deeper.

What you can build

Education products tend to be one of a few shapes, each on infrastructure that already exists.

A tutoring copilot that adapts to each student, remembering what they've covered, where they struggle, and how they prefer to learn, across sessions, the learning and tutoring copilots pattern: per-student Bases, persistent memory, searchable course material.

A lecture and course-content platform that transcribes lectures and makes them searchable by meaning, so a student can find the moment a concept was explained rather than rewatching, Extract plus Deep Search over course media.

A queryable course library where textbooks, notes, and recordings become searchable content a learner or agent can ask questions of, the document and media extraction patterns feeding a searchable store.

An adaptive-learning platform that personalises per learner at scale, with each student's progress isolated in their own environment, per-student Bases.

Education problems, solved

The problems education builders run into are specific, and each has an answer.

Personalisation that needs memory. A tutor that adapts has to remember the learner across sessions. Memory holds what a student knows, struggles with, and prefers, and keeps it current through contradiction resolution, so when a student masters something they once struggled with, the tutor stops drilling it. That's what makes the personalisation real rather than nominal.

Per-student isolation at scale. A platform has many learners, and each student's record must be their own. Bases make isolation structural, one per student, from a single API call, so individual learning records scale to thousands of learners without partitioning logic. It's the multi-tenant memory pattern.

Lectures locked in video and audio. Course content is often recordings nobody can search. Extract transcribes lectures with timestamps, so a student can jump to the moment a concept was explained, the same capability behind video and audio search.

Course material that's hard to search. Deep Search finds by meaning across textbooks, notes, and transcripts, so a student asking about a topic gets the relevant passage even when their phrasing differs from the material.

The infrastructure underneath

Four primitives carry most education products. Bases give per-student isolation from a single API call. Memory holds each learner's evolving picture across sessions. Extract transcribes lectures and turns course material into searchable text. Deep Search makes the course library queryable by meaning. Course material is typically shared across students while each student's progress stays private, the split covered in the learning and tutoring copilots use case. One API key for all of it.

Personalisation that deepens per learner

A learning platform on this foundation gets better for each student the longer they use it, which is the whole point of personalised learning. A copilot meeting a student for the first time has nothing to adapt on; after a term it knows their strengths, their recurring difficulties, and how they learn, and tailors accordingly. Because the memory self-organises, that accumulated understanding stays accurate as it grows, mastered topics marked as mastered rather than piling up as a confusing log. Per-student isolation scales from one API call, so the platform personalises for thousands of learners without the separation getting fragile, and a student who's built up a term of learning history has a real reason to stay, because that history is what makes the tool good for them specifically.

Get started

Start with the getting started guide, then the use-case pages that match what you're building: learning and tutoring copilots for the full pattern, multi-tenant memory for SaaS for per-student isolation, and video content search for lecture transcription. The AI flashcard generator example is a concrete build, and there's a free tier to build against.

FAQs

How does a tutor know what a student has already learned?

It remembers across sessions. Each session's interactions go to Memory scoped to that student, so on the next session the copilot knows what's mastered, what's shaky, and what to revisit, rather than starting from zero.

How are different students kept separate at scale?

Each student gets their own Base, a structurally isolated environment, so one learner's record never appears in another's session. It scales to thousands of learners from a single API call per student. It's the multi-tenant SaaS pattern.

Can it transcribe and search lectures?

Yes. Extract transcribes lecture video and audio with timestamps, and Deep Search makes it searchable by meaning, so a student can jump to the moment a concept was explained. It's the same capability behind video content search.

What happens when a student masters something they struggled with?

The new state supersedes the old through entity resolution, so a topic that moves from struggle to mastery is updated rather than kept as an open difficulty, and the tutor stops drilling it.

Does shared course material mix with a student's private progress?

No. Course material typically lives in a shared environment every copilot can search, while each student's progress lives in their own isolated Base. The split is covered in the learning and tutoring copilots use case.

Is this a finished learning product or something I build on?

Something you build on. Exabase is the infrastructure, per-student memory and isolation, transcription, and search, and you build the tutoring or learning platform on top. The learning and tutoring copilots use case shows the full build.