19 years across full-stack and AI/ML, based in Dubai. I build the plumbing under LLM-powered products — agent runtimes, trading systems, media pipelines.
Python, Rust, TypeScript, C/C++. LLM agent infrastructure · RAG systems · gRPC services · trading automation · GenAI media pipelines.
orno — terraform plan, but for agents
LLM agents fail silently and expensively. You learn what they did after they did it. orno pins a typed contract at runtime, so you read the plan, diff it, and reject it if it lies. Event-sourced, so replay and audit aren't a separate project. Built for CI from day one.
Rust GitHub Actions LLM Agents Event Sourcing
kavzi-trader — three agents walk into a futures market
Binance Futures automation on a Brain-Spine multi-agent setup. Scout scans, Analyst validates, Trader executes. Tiered LLM routing puts cheap models on the hot path and saves the expensive ones for decisions that actually move money. Every fill is event-sourced — no fill exists without a record of why it happened.
Python Pydantic-AI Binance API LLM Routing
claude-pipelines — agents all the way down
A plugin system for Claude Code: reusable multi-agent pipelines, skills, and hooks for day-to-day engineering work. The pipeline is built with agents and runs agents. Either elegant or a war crime, depending on the day.
Shell Python Claude Code MCP
cartoon-genai — a studio that fits on one GPU
End-to-end generative media on one workstation. FLUX.1 draws stills, Wan2.2 handles motion, ElevenLabs does voice, MusicGen writes the score, Claude writes the script. 9:16 video out, no render farm in.
Python FLUX.1 ElevenLabs MusicGen Claude
Earlier roles spanned full-stack systems, AI infrastructure, and voice/speech platforms. Same job, different stack each time: build the boring layer correctly so the product team above it can move fast.
Currently — shipping my own products at sgon.ai.
📬 Building something in AI and want to compare notes? → sgon.ai or drmozg@gmail.com