Claude agent systems
Skills, isolated subagents, hooks and MCP servers built with Claude Code and the Claude Agent SDK, so an agent can do real work inside clear limits.
I design and ship agent systems, RAG assistants and workflows that take repetitive work off your team, wired into the tools you already use.
Built with
I work with Claude Code, the Claude Agent SDK, MCP and n8n, then add the parts a demo skips: permissions enforced in code, retries, state you can inspect, and a human approval wherever a mistake would be expensive. One person designs it, builds it and answers for it.
About meThree systems that show the range: a governed Claude agent workflow, a client RAG assistant, and a live platform.
A local fund-administration prototype that delegates document review to Claude subagents while keeping permissions, exact calculations, drafting, and approval under explicit control.
A live CMS-led travel platform combining managed tours, inquiry capture, affiliate journeys, editorial content, operational email, and an MCP server for agent access.
A multilingual product and policy assistant that turns WooCommerce data and documents into grounded, inventory-aware answers.
Four kinds of work that usually end up connected in the same project. See all services →
Skills, isolated subagents, hooks and MCP servers built with Claude Code and the Claude Agent SDK, so an agent can do real work inside clear limits.
Workflows for documents, content, publishing and back-office tasks, with validation, retries and a queue someone can actually read.
Assistants that retrieve from your catalog, policies and documents before they answer, in French, English and Arabic.
Dashboards, APIs and sync layers that connect the automation to WordPress, WooCommerce, Supabase, your CMS and email.
Agent permissions are enforced with hooks and allowlists. A prompt asks for a boundary; code implements one.
Status fields, queues and retries mean a failed step waits to be recovered instead of quietly disappearing.
Agents prepare the work. Anything consequential stops at a clear human decision on the exact thing being approved.
The automation plugs into the store, CMS, database and spreadsheets already carrying the business.
Courses from Anthropic, Google Cloud and OpenAI sit alongside years of backend and web work, so the AI layer rests on solid data models, permissions and integrations.
Background and experienceWe map the current process, the tools involved, and where time is being lost.
You get a clear scope, a timeline, and a definition of done before any work starts.
I build in small complete pieces, test the critical paths, and show progress as it lands.
Documentation and a walkthrough, with optional monitoring and iteration afterwards.
Tell me what is slow, manual or fragile. I'll tell you honestly whether AI automation is the right fix.
Start a conversation