Review EnvironmentFrontend review build for institutional inspection.

Stevens IT Review

Stevens LTI Review Portal

Institutional review interface for architecture, controls, and synthetic stored embedding records.

Next.js App RouterStatic review build

Portal Navigation

Data FlowLaunch ValidationEmbedding Transposition

Launch-to-processing review path

The pre-registration review environment illustrates how a draft LTI launch would be validated, mapped to an internal learner UUID, converted into typed non-text events, and then transposed into fixed-length embeddings for downstream use.

Review Section

Pipeline overview

Each stage below reflects the documented pre-registration review path from draft platform launch simulation through typed-event conversion and onward to stored embeddings plus constrained metadata.

1

Canvas launch

The system models a pre-registration launch simulation using Stevens Canvas-aligned parameters and a draft validation path for issuer, deployment, token claims, and request integrity.

  • This review build does not imply an active Stevens Canvas registration or trust relationship.
  • The launch path is shown for architectural review before institutional registration.
2

Identity mapping

An opaque platform subject is mapped to an internal learner UUID in a restricted private table.

  • The learner UUID is the application-facing identity.
  • Identity linkage remains separated from downstream records.
3

Typed event conversion

Pseudonymized Canvas activity is converted into typed non-text events as an intermediate secure representation.

  • These typed events support secure preprocessing and policy controls.
  • They are not the persistent model-facing representation.
4

Embedding transposition

Each typed event is passed through an encoding layer that generates a fixed-length embedding plus constrained metadata.

  • Only embeddings plus constrained metadata proceed beyond the secure boundary.
  • Metadata includes learner UUID, event type, timestamp, model version, allowed-use tag, and provenance tag.
5

Database storage

Review architecture separates private mapping data, app data, and review-facing examples or summaries while storing only the downstream-safe representation for modeling.

  • Sensitive mapping data remains restricted.
  • Stored model-facing records are embeddings plus constrained metadata.
6

AI boundary

Downstream AI services see embeddings plus constrained metadata only and do not receive direct learner identity data.

  • Raw logs and unprocessed text do not persist for modeling.
  • Allowed-use and provenance tags remain attached to downstream-safe records.
7

Review visibility

The portal exposes synthetic examples of stored embeddings and constrained metadata, along with architecture summaries aligned to the review scope.

  • Synthetic records model the persistent downstream representation rather than the intermediate typed events.
  • Aggregate review outputs remain inspectable without exposing live data.
Typed non-text events are an intermediate secure representation. The persistent downstream representation is fixed-length embeddings plus constrained metadata rather than direct identity fields, raw logs, or unprocessed text.