Ghasi edTech
In development · pre-launch
- · AI-first by design
- · Offline-first by design
Multi-tenant, multilingual (LTR + RTL), AI-first, offline-first learning platform: marketplace + LMS + authoring suite + compliance engine. 18 bounded contexts. SCORM/xAPI. WebGPU local inference.
- Period
- 2025 — present
- Domain
- EdTech / Compliance
- Stack
- API Gateway · IAM
- Per-surface BFFs
- Search + Vector
- Event Bus
- AI Gateway · local + cloud
- Desktop · Mobile · Web
Modern learning platforms force a choice between cloud-only convenience and offline reach. Most also bolt AI on top of a non-AI core.
edTech is designed offline-first and AI-first from the domain layer up — not as features added to a connected system, but as first-class architectural posture. The same PlayPackage that streams in a browser is signed, encrypted, and runnable on a phone with no connection. Every AI-generated artifact — content, assessment, tutoring response — carries provenance metadata and routes through a single AI Gateway with model-selection, cost-ceiling, and moderation enforcement.
The layers, ports, and integration patterns are real — drawn from the live architecture baseline. Service names, identity providers, and external vendors are anonymized or labelled by role so the diagram reads as the posture, not as a deployment manifest. Hover any layer label on the left for context; right-edge annotations call out the load-bearing decisions.
— 01 —
One sync protocol (`/sync/v1/pull|push`) with idempotency and per-aggregate conflict policy.
CostEvery aggregate must declare its conflict rule.
WinA single mental model for every offline mutation.
— 02 —
All AI traffic through one AI Gateway service.
CostA service to operate.
WinModel choice, cost ceilings, provenance logging, moderation, and audit are centralized.
— 03 —
No synchronous cross-service chain longer than two hops.
CostMore events, more eventual consistency.
WinLatency budgets are tractable; every service can be reasoned about locally.
— 04 —
AI provenance metadata attached to every AI artifact — `{ model, version, promptId, traceId, reviewedBy, local }`.
CostA column.
WinEvery AI output is auditable and citable.
- 18 bounded contexts, 19 logical services.
- Full traceability matrix: requirements → capabilities → services → APIs → events → user stories → tests.
- Offline runtime ships SCORM 1.2/2004 + xAPI/cmi5 with the same PlayPackage on every platform.
- Pre-launch deployment in preparation for first content partner cohort.
Founder, lead architect, spec author, and primary implementer. The end-state specification (14 documents, 60+ epics) is mine — and so is the bulk of the implementation: domain code, unit and integration tests, Playwright E2E automation, CI/CD pipelines, and the cloud infrastructure underneath.
The authoring tool's block registry started as a discriminated union and should have been a versioned schema from day one. Migrating now is more expensive than it would have been.