For companies
Freelance AI engineer · Forward Deployed EngineerAI that actually works inside your company.
Freelance Forward Deployed Engineer: agents that act in your tools, automated workflows, assistants over your documents. Wired into your systems, measured, then handed to your teams. ENSEEIHT engineer, three-year work-study at Thales Alenia Space.
- of work-study at Thales Alenia Space: offline LLMs, encryption, audits
- 3years
- AI projects built, tested and measured (plus 3 channel projects: 19 on the Projects page)
- 16
- successful attacks on my agent: its security controls stopped all 14 attempts
- 0/14
- MCP connectors built and tested, to plug AI into existing tools
- 5
What I build
Five kinds of work
Each backed by a project that proves it: architecture, tests, measurements and documentation detailed on its page. Read access to the code under NDA, or a 30-minute live demo.
- 01
AI agents
Agents that handle a task end to end in your tools (support, back office, CRM), under a policy they cannot bypass, with human approval.
Tool calling · MCP · Stripe · HubSpot · FastAPI
The proof: RELAIS → - 02
Workflows and automation
Your repetitive tasks automated, monitored and documented: they resume on their own after an outage and never process the same item twice.
n8n · REST APIs · webhooks · MCP · SQL
The proof: Studio Flow → - 03
Assistants on your documents
Grounded answers over your contracts, procedures, technical documentation and internal knowledge, with your directory’s permissions, an audit log and an offline mode.
Hybrid RAG · OIDC · Ollama · encryption
The proof: Vigie → - 04
AI content pipelines
Your teams produce images and video at volume, on brand: ComfyUI workflows, a LoRA trained on your product, batch quality control.
ComfyUI · LoRA · ControlNet
The proof: RELIEF Studio → - 05
Models and prediction
Predict performance, classify, summarise: models trained or fine-tuned on your data, evaluated out of sample before they are used.
PyTorch · LoRA / PEFT · MLX · TimesFM
The proof: The Oracle →
Security and data
Your data stays with you
The first question teams ask, before any project. NDA signed before you share anything confidential, if you wish.
- 01
Where you decide
Your models can run on your servers, with the host of your choice or through an API, depending on your constraints.
- 02
Nothing leaves without consent
No data goes to an external service without your written consent.
- 03
Within your GDPR framework
Or the Swiss nFADP: data processing agreement, minimal data, access log.
- 04
Already done offline
At Thales Alenia Space I deployed and maintained LLMs on a fully offline infrastructure.
Engagement formats
Start small
A fixed-price assessment to decide, a pilot to prove, embedded support to go further. You can stop at any point.
A Assessment
2–3 days · fixed price
Workshops with your teams: priority use cases, data inventory, security constraints.
You getA report: priority use case, recommended architecture, risks, measurable success criterion, pilot budget.
B Pilot
3–4 weeks
An agent, workflow or assistant deployed on your infrastructure, on a real scope.
You getA running system, an evaluation set, before and after measurements.
C Embedded support and scale-up
Day rate
IT integration, connectors, scaling, security review of your AI projects.
You getVersioned code, documentation, runbook: your team takes over.
How I work
In the field
A Forward Deployed Engineer makes AI work at your company, not just in a prototype.
- 01
Immersion
I spend time with the people who do the work: their tools, their cases, their exceptions.
- 02
Prototype
A first version on your data within two weeks, to see quickly what works.
- 03
Measurement
An evaluation set built with your experts. Every version is measured, regressions are blocked.
- 04
Hand-over
Documentation, runbook, versioned code: your team runs and evolves the system without me.
Projects
Built and measured
Three examples run as client missions, with fictional clients. Each has its tests, measurements and documentation.
RELAIS
FunctionalA support agent that acts in your systems, under a policy it cannot get around
0 / 14 attacks succeeding with the gateway (14 / 14 without), even with an agent deliberately rigged for the test
Vigie
FunctionalSovereign, auditable RAG for confidential documents
0 leaks: evaluation 32 / 32 on SQLite and PostgreSQL alike
The Oracle
FunctionalShared data platform and forecasts evaluated over time
−21% error on 28-day views against the rolling median: log10 MAE 0.499 → 0.393 (102 test videos)
FAQ
Company questions
What teams ask before a mission.
Can my data stay on my premises?
Yes. At Thales Alenia Space I deployed and maintained LLMs on a fully offline infrastructure. Your models can run on your servers, with the hosting provider of your choice or through an API, depending on your constraints. No data leaves for an external service without your written consent, and I work within your GDPR (or Swiss nFADP) framework: processing agreement, minimal data, access log.
What is a Forward Deployed Engineer?
An engineer who works alongside your teams rather than from a ticket: they learn the business, wire AI into your real tools and data, measure the result and stay until it runs.
Do you work with our existing tools?
That is the point: your ERP, CRM, ticketing tool (Zendesk, Freshdesk…), email, directory and technical documentation, through their APIs, scheduled exports or database reads. I start read-only; any action that writes to your systems goes through human approval first. I adapt to your systems, not the other way round.
How do we know it works?
We build an evaluation set with your experts at the start, on your real cases. Every version is measured against it, and a version that regresses does not reach production.
After the pilot, who maintains the system?
Your team, with the documentation, runbook and tests delivered: the system is built to be taken over. If you prefer, I can also handle follow-up (updates, measurements, fixes), at a fixed price or day rate.
Who exactly are you?
An ENSEEIHT engineer (class of 2026), three years of work-study at Thales Alenia Space on offline, encrypted systems. I work alone, as a freelancer. The projects on this site are built and measured; RELAIS and Vigie are self-initiated projects run in a client format with fictional clients, as their pages state. Any engagement I start, I finish, and I deliver it so your team no longer needs me: code, tests, documentation and access live in your repository, not with me.
Where and how do you work?
Freelance, fixed price or day rate. On site in Toulouse, Paris and anywhere in France or Switzerland for workshops and launch, remotely the rest of the time. In French or English.