An execution machine showing how data flows through an AI agent

AI engineering & agent development

AI is no longer a concept. It's your execution engine.

Flowmetrik is an engineering firm specialised in agent development. Plenty of people will sell you a deck about artificial intelligence. We write your agents, prepare the data underneath them, and put AI into production — with founder Mehdi Bakkali.

01Data
02Agent
03APIs
04Production

They trust us with their data and their critical processes.

Flowmetrik is an engineering firm specialised in agent development. Plenty of people will sell you a deck about artificial intelligence. We write your agents, prepare the data underneath them, and put AI into production — with founder Mehdi Bakkali.

01Carrefour
02Agile Immo
03Consult'im
04BNP Real Estate
05Stratehab Conseil

Why Flowmetrik

70% of AI projects fail.
Not ours.

The reality check. 70% of corporate AI projects fail: endless proofs of concept, poorly structured data — theory is no longer enough.

Our paradigm. The era of the 10x developer: modern workflows (Cursor, Linear, automation) and robust data architectures. A small team ships in weeks what used to take months.

Our commitment. Not billable hours of advice: we build systems that run on their own, pay for themselves and scale.

The role we deploy

Forward deployed engineer

This is what separates a POC from a system that actually runs. A forward deployed engineer does not hand over a recommendation: they sit inside your organisation, with your data and your constraints, and write the code alongside the teams who will use it.

  1. 01On your ground, not in a demo environment.Your files, your access, your team meetings — messy exports and unwritten rules included.
  2. 02Code, not a slide.An engineer first. What is left behind runs, reads and operates without us.
  3. 03Autonomy is the exit criterion.We measure your teams' weekly usage, not contract renewal.

We are hiring forward deployed engineers. You code with agents, you learn a business before writing a line, and client ground appeals to you more than a meeting room: mehdi@flowmetrik.com

0113Agents deployed
022 500+Hours automated
0320+Systems integrated
0415Happy clients
Mehdi Bakkali, founder of Flowmetrik

Leadership

Mehdi Bakkali

Founder & CEO

An engineer (Polytechnique, INSA) with an HEC Paris MBA, Mehdi has led data and real estate projects in demanding environments — retail, banking, consulting. He takes leadership teams from proof of concept to production: bespoke agents, data you can trust, teams equipped to carry on.

View LinkedIn profile
Training
HEC ParisÉcole polytechniqueINSA
Companies
BNP ParibasSociété GénéraleCarrefour

Areas of expertise

What we actually do
three things.

Automate your tasks, make your data reliable, and leave your teams self-sufficient.

01

Business AI agents

An agent connected to your tools takes a request, follows your rules, and prepares or performs the task under your control.

Where the agent acts
GmailSlackNotionMicrosoftGitHub
02

AI-ready data

We gather, clean and connect your files, CRM and databases so that AI works on data you can trust.

The sources we bring together
SalesforceHubSpotMicrosoftGoogle DriveSAPPostgreSQLSnowflake
03

AI training on your own projects

We automate one real task using your own tools, then train the team to do it again without us.

The tools we put in its hands
ClaudeOpenAIGitHub CopilotMistral AI

What is an AI agent?

A brain given
arms and legs.

The model reasons. Memory gives it context. Skills tell it how to work. Connectors let it act inside your tools — under human control.

Input

User

A request through chat

Output

Answer or action

A result you can check

Context

Memory & RAG

History, rules, internal knowledge

Data

Files

PDFs, spreadsheets, documents, databases

The brain

AI agent

A model given memory, skills, and the means to act.

ReasonChooseRun
AnthropicGoogle GeminiMistral AIOpenAI
Skills

Skills & code

Business methods, scripts, controls

Extension

Internet & agents

Web search and agent-to-agent coordination

Arms and legs

APIs & tools over MCP

SalesforceSAPNotionSlackGoogle DriveGitHub

LLM The reasoning layer can change provider without rebuilding the system.

MCP Tools are exposed to the agent through standard, controlled connectors.

A2A Several agents can split a mission between them and share what they find.

Sector unit

A real business ground.
A discipline of execution.

FlowSpend focuses our engineering on one specific business system: contracts, invoices and supplier data. Same demands for proof and for production as everything else we build.

01

Spend & procurement

FlowSpend

We bring together contracts, invoices and supplier data to show where spend leaks and what can be recovered.

Contracts · invoices · supplier performance

Public reference
Carrefour

Achats non marchands

Explore FlowSpend

How it works

From idea to production,
with proof at every step.

The project moves through short decisions. Each phase delivers something verifiable before the next one is committed to.

01

3–5 jours

Frame

Business problem, data sources, users, risks and the metric that defines success.

Go / no-go decision
02

2–4 semaines

Prove

A prototype wired into a real scope, to measure the value before scaling it.

A working demonstrator
03

4–12 semaines

Deploy

Connectors, security, testing, observability and change management, inside your environment.

System in production
04

Continu

Evolve

Usage measurement, sharper skills, new workflows, and handover to your teams.

Compounding value

FlowStart frames the work and reports back.Flowmetrik designs, integrates and ships to production. The tool supports the work; it replaces neither the advice nor the delivery.

The Flowmetrik DNA

Our values.
Our code.

01

Transparence

No black boxes, no jargon to paper over emptiness. You know exactly what we build, how we build it, and why.

02

Efficiency

A cult of results: lean process, sharp tooling, zero friction. We go straight to the point to maximise your ROI.

03

Delivery

Execution beats theory. Our measure of success: what runs reliably and securely in your production.

The team

Technical excellence,
forged by high standards.

A modern, expert team, trained at France's leading business and engineering schools. Specialised, and used to delivering for large accounts.

Mehdi B.01

Mehdi B.

Founder & CEO

Leads Flowmetrik. Engineer and HEC Paris MBA — product vision, AI execution and the business case.

LinkedIn
Sami B.02

Sami B.

Software Engineer

Builds the integrations and software services that put agents, APIs and AI pipelines into production.

LinkedIn
Hadrien P.03

Hadrien P.

Project Manager

Runs projects, bridges business and engineering, and owns delivery — from roadmap through to production.

LinkedIn
Ines B.04

Ines B.

Head of Design

Design lead — experience, premium spaces, wellbeing and materials.

LinkedIn
Ines E.05

Ines E.

Head of AI & project management

Data & AI consulting — agentic AI initiatives, data governance and adoption at scale.

LinkedIn
Kamil B.06

Kamil B.

Senior AI Engineer

Model and agent architecture, Python stack, and shipping AI tools to production.

LinkedIn

Ready to go to production?

Let's assess your project.

Mehdi Bakkali — Flowmetrik · mehdi@flowmetrik.com

Go to production