mohamed gomaa

I’m a systems engineer designing production AI tools that prioritize architectural leverage over feature accumulation. I build and ship compact, self-improving agents that run on constrained hardware—local inference, CPU-first deployment, and offline-first workflows—without depending on cloud-only pipelines.

One of my flagship systems is BrowserMind, where I reframed “LLM per-step” browser automation as a bounded decision problem. I trained a 470K-parameter browser navigation model using Behavioral Cloning and DAgger without a GPU cluster, including constraint filtering, heuristic fallback, and decision-aware training loops that improve from live sessions.

I’m also building NotebookMG, a multimodal cognitive OS with offline-first RAG running locally. It’s a licensed desktop product architecture with a local inference stack (Ollama/LM Studio), a RAG vault, SRS and exam engines, and an AI companion—engineered for reliability and privacy rather than cost as a workaround.

In parallel, Interactive Cases focuses on clinical reasoning with 200+ AI-generated cases and decision-tree modeling in Arabic and English. As a medical student, I bring domain depth into information architecture and decision modeling, and I apply that rigor across my full lifecycle—from constraint definition to architecture, training loops, deployment, and licensing.