AI engineer. I build and operate production LLM systems.
Backend foundation → AI. I care about the unglamorous half: reliability, cost, guardrails, and making AI tooling actually work for the people who depend on it.
Recent finds: a CI job that failed 25 times in a row while printing ✅ Tests completed successfully,
and three weeks of a RAG pipeline running on random-number embeddings that still produced confident replies.
- 🤖 Production AI platforms — multi-tenant systems where LLM agents (WhatsApp, voice, document processing) run 24/7 for non-technical teams. I own architecture → reliability → cost.
- 🛡️ Making unreliable models safe — deterministic guardrails, instrumentation, incident response. Trust the model for fluency, never for invariants.
- 💸 Cost & performance — cut LLM inference cost 5.7× per document while holding accuracy steady, and built the telemetry to see real spend in the first place.
- 🧩 DSPy in production — optimizers, self-healing pipelines, and testing things that give a different answer every run.
- Document-AI platform — 54 document types, GPT-4o vision + DSPy, multi-tenant Azure, 438 tests in CI.
- Real-estate AI platform — solo-built: WhatsApp + voice sales agents, Microsoft 365 sync, auto-publishing, social/ads analytics.
- Supportly — RAG email assistant (DSPy + Qdrant + Gmail).
- Open source — a merged fix to Chutes, a Bittensor subnet project, plus an open data-integrity PR to Macrocosmos data-universe.
Python · FastAPI · DSPy · Claude / GPT-4o · PostgreSQL · Next.js / TypeScript · Azure · Docker · Qdrant
Based in Tunisia. Open to AI and backend engineering roles anywhere, remote or on site, and willing to relocate.
📫 bani.mahdi666@gmail.com · LinkedIn · X · Site