What I use.

Primary language for everything frontend and backend Node. Type safety catches mistakes at compile time — essential when the same data model spans API routes, UI components, and database schemas.

AI and data work. FastAPI for APIs, LangChain/LangGraph for agent orchestration, Locust for load testing. The ecosystem for AI tooling is unmatched.

Framework for all internal web apps. Server components let me keep sensitive data server-side without a separate API layer. Used for HR-OS and KPI Tracker.

For apps that don't need SSR — like the Interactive Code Lab embedded in the LMS. Vite's build times are under 3 seconds, which matters when iterating fast.

For lightweight servers and tooling. The Instructor Check-In system runs on Node with whatsapp-web.js — chosen because every instructor already has WhatsApp.

2.0 Flash / 2.5 Flash. Primary LLM for everything: content generation, exercise grading, multimodal chat, and TTS. The Egyptian Arabic quality and native multimodal support made it the right choice over alternatives.

Stateful conversation graphs with conditional routing. Used in Armstrong AI to handle text, image, and voice inputs through separate processing nodes that share the same session memory. LangChain chains couldn't model this cleanly.

Vector database for the RAG pipeline. Managed cloud — no infra to maintain. Cosine similarity search over Gemini Embedding 001 (768-dim) vectors for course content retrieval.

API layer for the AI system. Async by default, Pydantic validation, auto-generated OpenAPI docs. The /docs endpoint alone saves hours of debugging time.

Managed PostgreSQL with built-in Auth and row-level security. Used for HR-OS. RLS enforces data access at the database layer — not just the API — so even a buggy route can't leak cross-tenant data.

ORM for the KPI Tracker. Type-safe query builder catches join mistakes at compile time. The generated types flow end-to-end from database to API response without manual duplication.

Relational store for conversation history in Armstrong AI. pgvector handles the embedding store that doesn't need Qdrant's ANN performance — simple similarity queries on small sets.

Document store for the Interactive Code Lab. Write-heavy, user-isolated, real-time updates to the instructor dashboard. The document model fits per-user-per-exercise state perfectly.

All backend services are containerized. docker-compose bundles the app + PostgreSQL in one command. The KPI Tracker runs entirely in Docker on a VPS — one deploy.sh script handles pull, build, restart.

Hosting for Next.js apps and static sites. Zero-config deploys, edge network, automatic preview URLs. This site runs on Vercel.

Server hosting for the Check-In system. Persistent VM so the WhatsApp Web session stays alive between sessions.

DNS and proxy for mostafafathy.com. Proxied CNAME to Vercel, with automatic SSL and DDoS protection. The free tier handles everything this site needs.

CDN-backed static hosting for the Interactive Code Lab. Sub-second deploys from firebase deploy.

AI coding assistant for everything from architecture decisions to implementation. Used as a pair programmer that has full codebase context. This site was built with it.

Primary editor. Extensions: Prisma, ESLint, Tailwind CSS IntelliSense, GitLens, Docker.

Version control for all projects. Internal work lives in the Armstrong org. Personal showcase repos at MostafaFathy1527.

Design for all UI work. The Armstrong brand system and this site's visual identity were designed in Figma before any code was written.

API testing and documentation. Used heavily during the KPI Tracker's 51-endpoint development cycle.

Project management, sprint planning, and documentation for the E-Learning team.

Primary OS. WSL2 for Linux tooling when needed. PowerShell for scripting and deployment automation.