Yassine E.

I build production AI agents in TypeScript on Mastra: tools with strict schemas, multi-step workflows with retries, human approval in Slack, scheduled runs, tracing on every run, and cost controls that fail closed. 21 apps shipped, Top Rated Plus, 100% Job Success. Proof, not promises. I run an autonomous marketing team for my own app. Every morning at 06:00 a Mastra workflow researches, writes and lints a content pack for nine platforms and posts a Slack card for each item. Nothing publishes until a human presses Approve. Every model call is metered, a spend cap stops the run before it overspends, and a full daily pack costs about one cent. It runs on OpenAI and Gemini with LibSQL and DuckDB behind it, and about 170 automated tests cover it. What I take on: • AI agents and workflows in TypeScript: tool calling, multi-step workflows, human-in-the-loop approval, scheduled jobs, RAG over your data, Slack and Telegram surfaces, MCP servers and clients • Rescuing agents and automations that are slow, expensive or unreliable: I add tracing, evals, retries and a cost ceiling, then fix what the traces show • Adding AI to an existing app: LLM features, structured output, image generation, with a budget that does not surprise you • Mobile apps with AI inside: Expo or Flutter, Supabase or Firebase, App Store and Play Store release The mobile track is how I got here. Since 2019 I have shipped 21 client apps to the App Store and Play Store on Expo and React Native, Flutter and FlutterFlow, with Supabase, Firebase, NestJS and Next.js behind them: booking with Stripe Connect, marketplaces with encrypted chat, healthcare, restaurants, and AI features inside the app (OpenAI pantry scanning, Gemini story generation). My own app, Fouz, is live on iOS. Clean code, clear communication, no mess left behind. Arabic, French and English. Tell me what the agent should do and who has to approve it, and I will tell you how I would build it.