// data · ai · architecture · leadership

From data strategy
to production.

10 years of experience, including 7 years leading data & tech teams of up to 35 people. I help organizations design, industrialize and govern their data platforms and AI solutions — while keeping my hands in the code.

Let's talk about your project → View achievements
strategy vision · governance architecture data mesh · cloud delivery teams · dataops production ml · ai · measured impact
35team members managed, 4 units
€2Mannual budget owned
300+dbt models in production
20,000weekly field interventions optimized by ML
// expertise

Three ways to create value

Leadership, architecture or delivery: the scope adapts to your needs. In every case, I remain hands-on — from POC to production code.

◆

Data & AI Leadership

Department or program leadership: data-driven strategy, team structuring and upskilling, data governance, change management and alignment of metrics with business objectives.

Head of DataDAMA-DMBOK ADKAROKR Team TopologiesAgile/Scrum
⬡

Solution Architecture

Design of cloud data platforms and application solutions: Data Mesh, medallion architecture, real-time and event-driven integration, architecture documentation and cost / performance / security trade-offs.

Azure · GCPSnowflake dbtData Mesh Event-drivenC4 Model Well-Architected
⚙

AI & Automation

From ML models in production to AI agents: prediction, optimization, LLM & RAG, OCR, business process automation. Full industrialization — CI/CD, A/B testing, monitoring.

MLOpsScikit-learn AI AgentsLLM / RAG Vector DBFastAPI
// achievements

Clarify a problem, make the call, deliver

Six complex problems, solved from design to outcome. The architecture details live on a dedicated page.

01

Scaling FTTH field operations

Problem
Manual technician scheduling, unmanageable at 20,000 interventions/week — a 3-day repair lead time.
Diagnosis
The delay wasn't a shortage of technicians, but manual assignment and avoidable repeat visits.
Decision
Rather than a perfect prediction model, a simpler chain that was production-ready and adopted in the field.
Impact
Repair lead time cut from 3 days to 1-2 days, operational scheduling automated.
View the architecture →
02

Building one shared truth: the Data function

Problem
No data function: every team produced its own figures, with no shared truth to decide on.
Diagnosis
The problem wasn't reporting but the absence of a single source of truth.
Decision
Impose an SSOT approach from the start — politically harder, but the only path to a shared truth.
Impact
Analytics adopted by 5,000 users, 500+ reports, foundation of a 35-person department.
View the architecture →
03

Distributing €3M under legal constraint

Problem
Distribute a €3M variable-pay envelope across 3,500 employees, under regulatory and deadline constraints.
Diagnosis
The real risk wasn't the calculation but salary-data confidentiality under time pressure.
Decision
Security and deadline first: encryption and per-level entitlements, functional scope second.
Impact
€3M distributed on time, access partitioned and traced, compliant process.
View the architecture →
04

Launching an AI SaaS on a tight budget

Problem
Build an AI business and a SaaS product from scratch, with limited resources.
Diagnosis
Cloud costs and technology lock-in could kill the project before any traction.
Decision
A self-hosted, open-source infrastructure over managed services: control over cost and data.
Impact
SaaS in production, operational AI-agent architecture, autonomous team.
View the architecture →
05

Breaking the central data bottleneck

Problem
The central data team, handling all demand, was blocking the entire company.
Diagnosis
At that scale, centralization no longer held: domains had to become autonomous.
Decision
Evolution to a Data Mesh: domain ownership, central team refocused on pipelines.
Impact
From one to many owning teams, 300+ dbt models, self-service indicator design.
View the architecture →
06

Modernizing an opaque financial engine

Problem
A legacy financial consolidation engine, rigid and slow, with unknown calculation rules — under deadline.
Diagnosis
Before modernizing, the black box had to be reverse-documented to guarantee accuracy.
Decision
Reliability first, performance second: reverse-engineer the rules before any rewrite.
Impact
Engine in production, sub-minute latency, documented rules.
View the architecture →

What 10 years of data taught me

  1. The best decisions are made closest to the field and the users.
  2. Simplicity is the highest form of architecture.
  3. A solution only has value once it is truly adopted by the business.
  4. Data transformation rests on organization and governance more than on technology.
  5. High-performing teams are built on trust, autonomy and skills development.
  6. A project's success is measured by its lasting impact on the business.
// method

A pipeline, from need to impact

Every engagement follows the same logic as a data pipeline: clear, tested stages that only ship downstream what has been validated upstream.

01

Understand

Needs audit, field listening, mapping of the existing landscape and business pain points.

discovery · double diamond
02

Design

Target architecture, technology trade-offs, governance and migration trajectory.

c4 model · well-architected
03

Prove

Focused POC on the highest-value use case, with measurable success criteria. Output: a go/no-go decision — ROI, compliance, build vs buy.

mvp · a/b testing · go/no-go
04

Industrialize

CI/CD, testing, data quality, documentation: from prototype to reliable product.

dataops · mlops
05

Transfer

Team upskilling, rituals, autonomy — the value stays with you.

adkar · mentoring
// background

10 years across data, AI and leadership

Companies are anonymized — references and details available in direct conversation.

Data Architect / Data Engineer

International industrial group (electronics) · 2025 – 2026

Deliberate technical assignment: real-time financial consolidation engine, dbt/Snowflake pipelines, governance standards and coordination of external teams.

Data & AI Lead

Independent SaaS venture · 2025

End-to-end leadership of a data & AI activity: team of 6, AI-agent SaaS ERP, fully self-hosted infrastructure.

Head of Data & Software Development Department

National telecom operator · 2022 – 2024

Department of 35 people across 4 units, ~€2M budget. ML in production, smart applications, Data Mesh architecture and a unified group data platform.

Head of Data

National telecom operator · 2019 – 2022

Built the Data unit from scratch: team of 16, on-premise architecture (data lake, DWH, data marts), governance, data evangelization across operational units.

Data, BI & analytics engineering

Telecom, consulting & business travel · 2015 – 2019

Python/R/VBA development, report automation, business simulations, BI dashboards and analytics project management.

Master's degree — Organizational Sciences

Université Paris-Dauphine · MBA Erasmus exchange (Vienna)

Languages

French (bilingual) · Arabic (native) · English (professional)

// contact

A data or AI project in mind?

Available for new opportunities — data leadership, solution architecture, digital transformation or high-value AI solution design. Based in the Paris area, working across France and internationally.

salmen.khemir@gmail.com LinkedIn ↗