track b / best use of cognee cloud
Content apps give a class a syllabus. Cognee gives it a memory. Classroom Memory gives the teacher a plan. Each student gets a private Cognee Cloud memory; the teacher reads across all of them and gets told what to teach next, and why.
Pick a student and start a quiz. Every answer reshapes their knowledge graph, live.
The easy way to use a memory layer is one big graph. But a classroom is the opposite: every student is a private, isolated Cognee Cloud dataset, and the teacher reads across all of them. That is a multi-tenant problem, impossible on a single local instance. This is what the Cloud is for.
about the project
Classroom Memory gives every student an isolated Cognee Cloud memory and gives the teacher a live map of what the whole class is forgetting. It is built like a reusable classroom product, not a one-off demo: deterministic curriculum, enrollable students, cloud verification, and a clear production path.
In India and much of the world, one teacher faces 40 to 60 students. Personal attention is mathematically impossible, so gaps stay invisible until exams, and each year's gaps compound into the next. EdTech shipped videos and question banks: content at scale. What was never built is memory at scale: a system that knows what each individual child knows.
Hold each student's knowledge as a real memory: a graph of concepts with mastery weights that rise with correct answers, decay with neglect, and live in Cognee Cloud so they survive every device and every session.
An adaptive quiz that targets your frontier, an ask box answered from your own memory graph, and a teacher heat map that reads across every student's dataset to say what to teach next. Add a student and the same workflow starts from a blank memory.
functional architecture
The browser holds zero learning state. The backend scores mastery and walks the curriculum graph. SQLite keeps the exact classroom ledger, while Cognee Cloud stores long-term student memory, class memory, recall surfaces, learning traces, and teacher interventions.
Student mastery, quiz, ask box, report card, timeline, graph inspector, class ask, teaching plan, roster setup, curriculum import, and review assignment.
Validates requests, selects frontier concepts, scores quiz answers, builds report data, ranks graph-reasoned teaching plans, and keeps demo/cloud modes behind one response shape.
Durable exact state for the classroom: students, concept weights, update timestamps, retired flags, seeded datasets, tenant metadata, and intervention assignments.
Cognee stores private student datasets, a dense class graph, session-memory quiz attempts, progress snapshots, report-card recall context, and teacher intervention memory.
Students see what to learn next and why. Teachers see the class heat map, a graph-reasoned lesson plan, targeted review lists, and per-student drill-down.
class_graph or fall back to multi-dataset recall across student memoriesforget(dataset)why cognee cloud
Isolation per student plus a teacher reading across datasets is a multi-tenant hosted memory problem. A local, single-user instance cannot build this view.
Seeds each student's dataset with the curriculum and writes a trace every time a concept is mastered.
~9s ingest, verified live
The ask box answers from the student's own cloud graph, with dataset provenance attached.
~7s warm, verified live
Mastery follows Cognee's feedback weight design, applied app side. Cloud does not expose remote improve() yet; the provider boundary is ready for it.
feedback_weight semantics
Retire a mastered concept, or reset a transfer student with a real dataset deletion on the tenant.
~2s delete, verified live
not a staged demo
This is not just three hardcoded students and a pretty graph. It has the minimum product spine a school would need: enrollment, isolated memories, repeatable curriculum files, deterministic assessment, teacher aggregation, cloud verification, and a clean path to production storage and auth.
Add a student with one click. Their memory is a brand new Cognee Cloud dataset, created on their first question and readable from any device, forever.
+ enroll
The Python course is a plain JSON file: concepts, prerequisite edges, questions, and answer keys. Adding a new subject is adding a file, not writing code.
curriculum/python.json
A one-command script asserts the full memory lifecycle against the live tenant, with timings, and exits non-zero if anything is fake.
python backend/verify_cloud.py
The current app is useful for pilots. The next production steps are straightforward: database-backed mastery ledger, school auth, curriculum upload, and native Cloud improve when exposed.
pilot today -> production next
execution quality
The project separates product behavior from cloud plumbing, which is what makes it credible beyond a hackathon screen recording.
DemoProvider and CloudProvider return identical JSON. The frontend does not care whether it is offline or backed by Cognee Cloud.
provider pattern
The teacher view is not a dashboard pasted on top. It reads across isolated student memories with one multi-dataset recall path.
datasets=[student_*]
The demo-mode smoke test covers quiz progression, decay, retire/reset, class ask, enrollment, and invalid student guards.
backend/test_demo.py
The site names what is cloud-native today and what is planned. That makes the project stronger, not weaker: judges can trust the claims.
no overclaiming
cloud console proof
For the final video, show Cognee Cloud itself, not only this app. Open the tenant console and capture the datasets plus the latest activity from curriculum seeding, mastery traces, and teacher interventions.
Show `student_alice`, `student_bob`, and `student_cara` in the Cloud console. These prove the per-student isolation boundary.
dataset per student
Open one student dataset and show the curriculum memory seeded with `remember()` and NodeSet tags.
curriculum node_set
Master a concept in the quiz, then show the new trace activity written into that student's dataset.
mastery-trace
Assign review from the teacher view and show the `class_interventions` dataset entry.
teacher-intervention
demo flow