AI engineering: agents, memory and MCP in production
How I build AI-agentic systems that do real work — agent memory, retrieval, MCP tooling and agents in production — with the products and engagements behind it.
The hard part of an AI agent is rarely the model call. It is everything around it: what the agent remembers between sessions, how it finds the right context, which tools it can reach and how, and what stops it from doing the wrong thing confidently. That is the work I focus on.
I built CaBrain because agents without memory start from zero every time, and Orchestra MCP because the tools an agent uses should be plugins, not a hard-coded list. Professionally I delivered an Arabic-language AI intelligence platform of analyst agents at ID8 Media, and designed an agentic operating layer — agents, evals, guardrails, memory and observability — at One Studio.
Products
- CaBrain — memory infrastructure for AI agentsPersistent memory, hybrid retrieval and an entity graph, over MCP.
- Orchestra MCP — plugin-based infrastructure for agentic developer environments290+ MCP tools as self-registering plugins, across IDEs.
- ToGO — a Go + React framework for AI-native applicationsEvery app ships agent-ready, with an MCP server pre-wired.
In production, for others
Roles where AI agents were the product or the operating model, from the experience record.