About
Nathan Guihot, independent AI specialist
I design, harden and ship AI systems to production, one point of contact from audit to run. Here is how I work, and what I do not do.
Who I am
I am Nathan Guihot. Under the guinat brand, I help businesses ship AI to production: agents, RAG, automation and workflows that hold up once delivered, not just in a demo.
I work solo and direct. You talk to the person who designs the architecture and writes the code, not to a salesperson and then a team you never meet.
What I actually deliver
Six areas, all production-focused, from idea to a system in use.
- AI agents and orchestration, plugged into your tools via MCP
- RAG and internal copilots, sourced answers, permissions respected
- Automation and workflows (n8n, custom integrations)
- EU cloud and on-prem, private models, sovereignty
- AI audit and LLM cost reduction
- Evals, guardrails, observability, red-teaming
How I work
Production first. The difference between a demo and a system in use is what you put around the model: evals that measure quality before every release, guardrails, observability.
Anti-hype. I tell you when AI would not pay off for you, and I would rather say no after a few days of audit than leave you disappointed after months of build.
One point of contact, from audit to run, with a documented handover so your teams keep control.
My principles
Your data stays yours. Depending on your constraints, everything can run on an EU cloud or inside your own infrastructure, with private models.
Measure instead of believe. Cost, quality, drift: what is not measured is not managed.
No vendor lock-in. An MCP connection is written once and works with Claude, GPT or a private model, so you stay free to switch.
What I have shipped
A few pieces of work delivered on client engagements, described without naming a client or inventing a number: an MCP infrastructure that gives agents controlled access to internal tools, an anomaly-detection system explained by agents, a RAG knowledge base kept in sync with the code, a log-ingestion filter that flags anomalies on the fly.
The full context is better told on a call.
You work solo, is that a risk?
Solo is a choice, not a limit: you get one point of contact, a standard and documented architecture, and a handover in every build. If extra hands are needed, I scope it and stay accountable for delivery.
Where do you work?
In France and across the European Union, in French and English, mostly remotely.
Where do we start?
A few-day audit, or straight to a free, no-obligation first call to scope your use case.
Contact
Ready to go from demo to production?
Reply within 24 hours · first conversation free, no strings attached.