AI Architect

I help companies of 20 to 500 employees rebuild their organization around AI.

95 agents in production at my own companies. Measurable ROI in 3 months.

I still run my 3 companies, but I only spend one day a week on them now.
Alexis Prat

Two profiles in one.

AI-First transformation requires two skills that rarely coexist: business judgment and technical depth. That profile barely exists in France — and is rare anywhere.

In 2025, I stepped out of operations for a year to master AI hands-on: building models from scratch, engineering RAG memory, and intensive experimentation.

01

Entrepreneur

CEO of 3 companies founded since 2017. 20M EUR bootstrapped revenue. 50+ employees hired. I advise from what I've lived, not what I've read.

02

AI Architect

I designed and shipped an 'AI-First' architecture across my 3 companies, including a persistent memory infrastructure that runs 95 autonomous agents today.

Free diagnostic

Where does your organization stand on AI?

10 questions. 5 minutes. A personalized recommendation sent by email.

  • 01Positioning across 4 levels of AI maturity
  • 023 priority actions for your context
  • 03Written report sent immediately
Start the quizFree · 5 min · zero commitment

IN PRODUCTION - LIVE

The AI agents dashboard.

The agents I've shipped, right now.

Active agents
95
multi-clients
Tasks / 24h
2 847
+12% vs last week
Cost per task
0,12 EUR
-18% vs Jan
  • 14:22:07agent-sdrB2B servicesqualified 12 leads, booked 3 meetings
  • 14:21:44agent-devismanufacturinggenerated a sales proposal in 18 min
  • 14:20:18agent-seodistributionpublished 3 articles (high-intent queries)
  • 14:19:52agent-financelogisticsreconciled 147 entries, flagged 3 anomalies
  • 14:18:33agent-crmB2B servicesenriched 87 contacts via Apollo + HubSpot
  • 14:17:09agent-projetconsultingclosed 2 deliverables and notified clients
Example inspired by real deployments.

The 5-layer architecture.

The architecture works as a stack. Each layer builds on the next. Everything is custom-built.

  1. L01

    Corporate memory

    Operational knowledge captured in .md files: processes, offers, personas, rules, history.

    Foundation
  2. L02

    Specialized agents

    Architects, analysts, executors. Each one has a mission. All read the memory. All document.

    Engine
  3. L03

    Stack integrations

    Slack · Gmail · Drive · HubSpot · Figma · Google Ads · Search Console. All your tools.

    Connections
  4. L04

    Supervised autonomy

    Agents act on their own. They escalate to a human for trade-offs or report sign-off.

    Operations
  5. L05

    Learning loops

    Every agent writes its lessons after each run and reads them at the next one. The system learns. This is the layer that makes agents better over time.

    Learning
The 5 layers are universal. Their content is unique to each company. Your operational memory, your agents, your stack, your governance: nothing is copy-pasted from another client. Anyone selling "off-the-shelf" AI workflows hasn't understood the problem.

Pick your format.

Format 01 / Entry

AI Audit

2 weeks. Mapping of your workflows, identification of 3-5 high-ROI projects, business case, prioritized roadmap. Deliverable: 15-20 page report + 2h debrief.

6,500 EURfixed fee

Deductible if Mission signed within 30 days

Recommended

Format 02 / Core

Transformation Mission

3 months. From audit to AI agents in production. I rebuild a core function of your organization around AI, then transfer to your team.

From 40,000 EURper mission

Typical ROI: 4-6x over 12 months

Format 03 / Ongoing

Fractional Chief AI Officer

Your Chief AI Officer without the full-time salary. 2 days/month, I govern your agents, deploy 1-3 new ones, and attend your executive committees. 6-month minimum.

10,000 EURper month

Format 04 / Keynote

Keynote & Masterclass

For conferences, executive committees, fund offsites. Concrete lessons on what works, and what doesn't.

2-5K EURper session

What it actually delivers.

Internal proof from my own group. First external missions are underway — client cases will land here as they sign off.

Albus Factory9 people

Sales, growth, ops, project

In 6 weeks, I avoided 4 hires. 9 people now do the work of 50. Less than 50,000 EUR invested. This is the live lab of the method.

  • 95agents shipped
  • 4hires avoided
  • 6 weeksto build
Boomrang Events11 people

MICE pipeline · brief → PPTX proposal

Second company in the group, same method. Autonomous sales pipeline shipped in 2 weeks. Proof of reproducibility: memory is unique to each company, the architecture is universal.

  • < 2hbrief → MICE proposal
  • 5h → 18 minaverage proposal time
  • 2 weeksto redeploy
Typical SME case20-500 employees

B2B services · distribution · logistics

This third row isn't a client testimonial — it's the kind of mission I run today. Outcomes depend on context; the method is identical to what's deployed at Albus and Boomrang.

  • 50-150Kmission (EUR)
  • 3 monthsto deploy
  • 4-6×typical 12-mo ROI

Latest articles.

All articles->

A 30-minute diagnostic.
Zero commitment.