Faheem AI — adaptive learning for thousands of Egyptian students

Situation

Appenza Studio was building Faheem AI — an adaptive learning app published by the Egyptian Ministry of Education and awarded a spot on the HolonIQ MENA EdTech 50. The platform needed to let students ask curriculum questions and receive grounded, curriculum-accurate AI responses, supported by audio narration and a monetization layer to enable school-by-school rollouts.

Task

As primary backend developer from early development, I owned and shipped several systems end-to-end:

• Build a 5-state adaptive learning chat with session-scoped AI and curriculum-grounded responses
• Design and implement a complete subscription and voucher system from business requirement to production
• Integrate platform-wide audio narration across all content types
• Build a cross-database sync service to keep two independent production databases consistent
• Manage Azure infrastructure across three environments

Action

Adaptive learning chat. I designed a 5-state conversation state machine persisted per session in SQL Server. Each session maintains its own conversation history, passed to Azure OpenAI on every turn. Responses are grounded in the official curriculum via RAG retrieval against Azure AI Search — the student always gets curriculum-accurate answers, not generic LLM output.

Subscription and voucher system. Built the entire billing module independently from scratch: tiered subscription plans with bundle and subject grouping, semester-scoped one-time voucher codes with redemption logic, role-based content access control, and the admin panel to manage it all. Designed the business logic directly from domain understanding — no technical guidance, no existing reference. This unlocked the school-by-school distribution the business team needed.

Platform-wide audio narration. Integrated Azure Text-to-Speech for audio generation and Azure Blob Storage for delivery. Refactored approximately 20 API endpoints across lessons, units, and capsules to support audio playback — every content type in the platform now speaks.

Cross-system sync service. Built a scheduled console app bridging Faheem (SQL Server) and the Question Bank platform (MySQL). Bidirectional ID mapping keeps content references consistent across two independently deployed production databases — content changes in one system propagate reliably to the other.

Backend migration contribution. Ported backend modules from Python Flask to .NET 8 as part of a broader platform migration, applying Clean Architecture and Mapster. Separately, moved AI prompt configuration from hardcoded Python files to the database in the voice-to-voice aiohttp proxy — eliminating redeployments every time prompt wording changed.

Azure infrastructure. Managed Azure resources via Azure CLI across DEV, STAG, and PROD — Container Apps, ACR, Blob Storage, Redis, PostgreSQL. Owned IIS deployments for multiple backend services.

Mentoring. Designed the image and voice upload feature architecture end-to-end, mentored 2 junior developers through full implementation, and reviewed all pull requests before merging to production.

Result

• Ministry of Education certification — the app is officially published on the Egyptian MoE platform
• HolonIQ MENA EdTech 50 recognition for the platform
• ~10K students in pilot using AI chat for real curriculum questions daily
• Subscription and voucher system enabled school-level rollouts the business team had been blocked on
• Platform-wide audio narration shipped across all lesson types — from zero to full coverage
• Two production databases staying consistent without manual intervention

What I’d do differently

Starting today I’d add structured evaluation infrastructure (golden question sets + automated scoring) from day one so prompt and retrieval changes could be tested before hitting production. The current setup works, but changes require manual spot-checking.