CSIntel Research Assistant
A multi-agent research tool built on the Claude Agent SDK — a parent orchestrator spawns specialised subagents that fan out, gather, and synthesise. The original LOG 001, and the system this site is named for.
CSIntel is the working portfolio of Sean Curgenven — a UK builder shipping Claude-powered tools, multi-agent research systems, and workflow automation. Every project is built, tested, and documented in public.
02.1 Currently building
Vision-model ingredient recognition in testing — photograph what's in the fridge, get a tailored recipe back. Claude's vision API doing the heavy lifting.
03 About the operator
I'm Sean — 24, based in Congleton in the UK, mid-transition from Energy Analyst to full-time builder. The day job taught me to read systems, chase anomalies, and trust the data. Now I point that same instinct at what modern AI can actually do when you stop demoing and start shipping.
CSIntel started as a personal lab and turned into the place where ideas get built for real. The stack is genuine — Claude API, Python, React, TypeScript — and the work ranges from vision tools to multi-agent research systems. No mockups standing in for products.
I build in public and document the process as I go — what worked, what broke, and what I'd do differently. The build log matters as much as the build. If you're here, you're reading it.
04 Selected projects
A selection of what's shipped, live, and in testing. The full index lives on the tools sheet.
A multi-agent research tool built on the Claude Agent SDK — a parent orchestrator spawns specialised subagents that fan out, gather, and synthesise. The original LOG 001, and the system this site is named for.
A restaurant AI assistant demo — handles menu questions, bookings, and opening hours in a fully branded site. A worked example of an AI concierge dropped into a real small-business front end.
Photograph your fridge, get a tailored recipe back. Claude's vision API identifies ingredients, handles gaps intelligently, and adapts to dietary preferences. Mobile-first for single-handed use.
05 Method
Find a real problem or a use case worth pressure-testing — something that reveals what AI can genuinely do when you stop demoing and start solving.
Use Claude and the sharpest available tools to ship fast. Iterate in the open. Don't wait for perfect — get something working and learn from what it does.
Log what worked, what didn't, and what's next. The build notes matter as much as the build — it's how the next project starts ahead of the last one.
06 Open channel
Freelance builds, API integrations, AI consulting, or just a conversation about what's possible — the channel's open.