LiberiaLearn：面向国家规模部署的 AI 教育平台

LiberiaLearn — Production AI Education Platform

🚀 AI-powered platform designed for national-scale deployment (1M+ students, 5,000+ schools)

What This Is

LiberiaLearn is a full-stack, production-grade AI system — not a demo — built to deliver real-world education infrastructure.

Built and operated by a single engineer (end-to-end system ownership), the platform is designed to function as national education infrastructure.

It combines:

• RAG-based AI tutoring (grounded responses, not hallucinations)
• Multi-tenant SaaS architecture (students, teachers, admins, MOE)
• Offline-first learning system for low-connectivity environments
• AWS-backed distributed infrastructure
• AI-driven workflows for tutoring, grading, and curriculum generation

Live System

👉 https://liberia-learn.vercel.app

Source Code

👉 https://github.com/fasiryon/liberia-learn

Key Metrics

• Tests: 1,577 passing
• Test files: 214
• API routes: 189
• Prisma models: 81
• Curriculum: 1,306 READY lessons
• Avg lesson length: 1,450 words

Traction

• 9,000+ repository interactions in recent weeks
• Increasing developer and system-level interest

Technical Highlights

• Multi-tenant role system (Student, Teacher, Guardian, Admin, MOE)
• Full App Router platform with 189 route handlers
• Prisma domain model with 81 models (users, curriculum, AI usage, governance)
• RAG-based AI architecture with retrieval + grounding + cost controls
• Offline-first system with IndexedDB sync + resumable sessions
• Immutable audit logging and governance reporting
• SLO tracking, health monitoring, and incident-ready architecture
• Full curriculum system with 1306/1306 lessons READY

AI System Design

• Retrieval: pgvector cosine similarity over curriculum chunks
• Grounding: responses constrained strictly to retrieved context
• Providers: OpenAI (primary), Groq (fallback routing)
• Prompting: structured templates with role, context, and curriculum injection
• Evaluation: recall@k and precision@k used for internal validation
• Cost Control: per-request tracking, rate limiting, and usage constraints

Engineering Decisions

• RAG used instead of fine-tuning to ensure explainability and curriculum grounding
• Multi-provider routing implemented to prevent vendor lock-in and increase reliability
• Offline-first architecture chosen due to unreliable connectivity environments
• Distributed workers (ECS + SQS) used for async scalability and background processing
• Observability built into core system for production readiness (CloudWatch + logs)
• Multi-tenant RBAC enforced at both API and middleware layers

Architecture

Repository Structure

• app → UI + API route handlers
• components → reusable UI components
• lib → auth, AI orchestration, services, telemetry
• prisma → schema, migrations, seeds
• worker → background job processing
• scripts → audits, pipelines, maintenance
• docs → architecture, governance, operations
• tests → Vitest suites
• infra → deployment and infrastructure

Local Setup

npm install
npx prisma generate
npx tsc --noEmit
npx vitest run
npm run build
___
## Environment 
Typical files:
	•	.env
	•	.env.local
	•	.env.production

See ENVIRONMENTS.md for full details.

___
## Documentation
Document
Purpose
ARCHITECTURE_EXECUTIVE.md
Technical overview
API_REFERENCE.md
API documentation
MOE_TECHNICAL_BRIEF.md
Ministry-facing explanation
SYSTEM_ARCHITECTURE.md
Deep system design
SECURITY_MODEL.md
Security controls
SCALE_READINESS.md
Scalability analysis
INCIDENT_RESPONSE.md
Incident handling
WORKER_DEPLOYMENT.md
Background worker setup

___
## Current Status
	•	Sprint 8 complete
	•	Sprint 9 documentation in progress
	•	Build validated via:
	•	npx tsc --noEmit
	•	npx vitest run
	•	npm run build
	•	Curriculum: 1306/1306 lessons READY

Execution tracking lives in [CURRENT_EXECUTION_STATE.md](C:/Users/fasir/liberia-learn/docs/roadmaps/CURRENT_EXECUTION_STATE.md).