Engineering the Al-Salem Educational Platform & Integrated Multi-Tenant Gemini AI Agent
Platform Summary
Al-Salem Educational Platform (alssalem.com) is a prominent Saudi EdTech examination and training platform serving thousands of secondary students preparing for Qudurat and Tahsili exams. To elevate interactive learning, a standalone multi-tenant AI Educational Agent microservice was engineered to act as an instant personal tutor, breaking down examination mistakes and delivering bespoke explanations.
The Engineering Challenges
1. Zero-Hallucination Guardrails
Educational exams demand 100% factual fidelity; the AI must strictly explain verified model answer keys without inventing incorrect steps.
2. Token Cost & Latency Optimization
Thousands of students concurrently taking exams can deplete API token budgets and incur response delays without intelligent caching.
Architectural Solutions & Engineering Highlights
- Scalable Multi-Tenant AI Microservice: Engineered an independent SaaS microservice using Node.js, Express, TypeScript, and Prisma ORM with tenant data isolation and platform API key authentication.
- Semantic & In-Memory Caching (node-cache + MariaDB): Slashed token consumption to 0% for repeated student queries, achieving sub-second responses and enforcing daily rate limits per student.
- Strict Educational Guardrails: Implemented proactive prompt filtering to keep student queries strictly bounded within accredited curriculum syllabi.
- Complex Multi-Role Frontends (Angular 22): Built centralized control suites serving 9+ distinct user roles (admin, supervisors, teachers, parents, accountants) with instant ExcelJS report generators.
Technologies Used
Angular 22
Angular Signals
TypeScript
Tailwind CSS
Node.js
Express.js
Prisma ORM
MariaDB / MySQL
Google Gemini API
Server-Sent Events