Freelance AI consultant: independent, honest, production-first

Why an independent, not an agency or a consultancy

With an agency or a large consultancy, you pay for a layer of account managers, project leads and subcontracting before anyone touches the code. With an independent AI consultant, you talk to the person who builds. One head, one number, zero telephone game between your need and what ships.

Solo doesn't mean fragile. I document the architecture, hand over code your team or another provider can pick up, and I lock nothing in: no dependence on a homemade tool, no black box. MCP is a good example: you connect your tools once, and it stays reusable with Claude, GPT or a private model, without me.

• One contact: the person who scopes is the person who codes.
• No sales layer and no subcontracting margin on your invoice.
• Documented code, transferable architecture, no lock-in: you stay free.
• Fast decisions: no committee to change a line.

What I work on

Six areas, one principle: production first, never the demo. Each has its own detailed page, where I explain the how and the guardrails.

• LLM & RAG: your documents become a base you query in plain language, every answer sourced.
• Agents & orchestration: agents connected via MCP that act in your tools, not chatbots that talk.
• Automation & workflows: your repetitive tasks handled by reliable n8n workflows, with alerts when something jams.
• Cloud & on-prem: your systems deployed wherever your constraints require, right down to your own infrastructure.
• Audit & optimization: I cost the profitable use cases and cut your LLM bill by 30 to 70%, at equal quality.
• Reliability & security: evals, guardrails, observability and red-teaming (OWASP LLM), for AI that doesn't go off the rails.

How I work: audit, POC, build, run

I never start with code. We move in stages, and each one can be a stopping point: you're never locked into the next.

At every stage you keep the code, the docs and the freedom to continue with me, in-house, or with someone else.

• Audit: I look at your processes AND the state of your data, and cost the profitable cases. First feedback within 48h.
• POC: I prove the value on a narrow scope, in a few weeks, with success criteria set in advance.
• Build: I build the system in production, with evals, guardrails and observability built in, not bolted on later.
• Run: I monitor quality and cost, and hand over everything your team needs to take over.

Systems in production, not slides

I talk about what I've built, without naming a client or inventing a number. A few examples of work actually delivered on assignment:

On results, I only quote orders of magnitude observed on comparable engagements, always in context: a sourced answer in 2 seconds and 70% less search time in consulting and legal; 10x more documents processed with no rekeying in insurance and finance; support available 24/7 with 50% fewer tickets escalated in e-commerce; 30 to 70% saved on an audited LLM bill, at equal quality. Your own numbers we set at scoping, not in a brochure.

• An MCP infrastructure that gives agents access to internal tools without opening a hole: a Zero Trust proxy in front of each tool, SSO, secrets never exposed, every LLM call observed.
• A three-engine anomaly detection system (statistical, machine learning, hybrid) across multi-OS fleets, with LangGraph agents that explain and prioritize the alerts.
• An end-to-end RAG pipeline feeding an internal assistant, continuously synced with the code repositories so it never drifts from reality.
• A log-ingestion filter that calls an LLM on the fly and returns anomalies as structured JSON, caught on the way in rather than after the fact.

Based in France, working across France and the EU

I work remote by default, which keeps me available everywhere: a freelance AI consultant in Paris, in Lyon or anywhere in France, and engagements across the whole European Union. Key sessions happen over video or on site when the project calls for it, especially scoping workshops.

Working from the EU matters for your data too: hosting on a European cloud or directly in your own infrastructure, outside the US Cloud Act when sovereignty requires it. Your data stays yours, wherever I am.

How I estimate an engagement: rate and scoping

I don't publish a price grid, because a number pulled from a table means nothing without your context. Cost depends on scope, the state of your data and how demanding production has to be. What I can tell you on the first call: an honest ballpark, and whether it's worth doing at all.

In practice, two formats. Fixed price for a scoped project, when the perimeter is clear: you know what you pay, I carry the overrun risk. Time and materials, at a daily rate, for ongoing support or a run, when the need keeps evolving. We pick the format that protects you best, not the one that suits me.

And the initial audit stands on its own: even if you don't build with me afterwards, the report and the roadmap are yours.