Freelancer profile translated to English.

Description

Deploying an AI system in 2026 involves simultaneously managing four risks that most POCs overlook:

Exposure of sensitive data to the LLM,
Dependence on a single vendor (hyperscaler or model),
Lack of traceability of decisions for AI Act audit purposes,
Uncontrolled inference costs at scale.

It is on these four axes that I have structured my interventions for 9 years, on Data/AI projects in production.

The method is consistent:

Agents in LangGraph
Vector RAG with NER anonymization upstream of the LLM
Orchestrated multi-agents
Hosted on Scaleway and/or any other French or European provider depending on the context
LangSmith or LangFuse for traceability
Token budgets for cost control
Model fallback for continuity
Queryable decision log for compliance

Context: I have co-managed DPLIANCE since 2019, a publisher of sovereign Data & AI solutions based in Poitiers. The studio publishes Mirage Analytics, Complio, and Cookilio, and delivers tailor-made AI solutions hosted in Europe. Our GDPR compliance software is used by several hundred organizations in France.

Recent references:

Winner of the ADEME 2024 Data/AI Challenge — autonomous agent connected to the SIRENE API to verify eligibility for public aid.
As part of an international team, we won a major account tender for a €70M PoC (communication exclusively in English).
Visiting lecturer in AI, AI Cybersecurity, and MLOps at Engineering Schools and Universities (ESIGELEC, IUT La Rochelle).

Three intervention formats:

AI Act risk audit for an AI system in production.
Design and production deployment of an agentic system from A to Z.
Fractional Chief AI Officer to frame an AI strategy.

If any of these issues are relevant to you now, two lines are enough to start the discussion.

hour.

Industry field of expertise

Languages

French
Native or bilingual
English
Fluent

Workplace preferences

Remote only

Primarily works remotely

ESIGELEC
Lecturer - AI Deployment, MLOps, and Cybersecurity
EDUCATION AND E-LEARNING
November 2025 - December 2025 (1 month)
Poitiers, France
Teaching a 30-hour course as a visiting lecturer coveringSecure Deployment of AI Solutionsin a production environment for final-year Engineering students.

Full Program:
Cloud deployment (AWS, GCP, Azure)
Robust MLOps
AI-specific cybersecurity and real-world use cases

Mixed assessment: continuous assessment (50%) and formal examination (50%).
Pedagogy Training Cloud Architect Cybersecurity awareness artificial intelligence
Nodac
AI Platform for LinkedIn Prospect Intelligence
SOFTWARE PUBLISHING
May 2025 - August 2025 (3 months)
Poitiers, France
AI & Infrastructure Integration:
LLM implementation via GroqCloud (data privacy, open-source models, scalable architecture)
Production monitoring with LangSmith (request traceability, performance, costs)
LangChain orchestration for LLM call management (modularity, scalability)
Security & Compliance:
Anonymization via NER (Named Entity Recognition) before LLM processing (GDPR-compliance)
Algorithmic Optimization:
Migration of clustering algorithm: K-Means → DBSCAN
Generative AI Machine learning Python GDPR IT Architecture
Data Retail Business
Design & Architecture of an AI Multi-Agent System
REAL ESTATE
April 2025 - July 2025 (3 months)
Poitiers, France
Strategic & Architecture Phase
In-depth analysis: Evaluation of the business problem and identification of optimizable workflows via multi-agent AI
Architecture design: Design of a distributed orchestration of specialized agents (hierarchical model with central orchestrator)
Technical stack: Selection and integration of tools (OpenManus, orchestration frameworks, cloud APIs)
Technical Implementation Phase
Development of the multi-agent system: Specialized agents with clear roles, collaboration mechanisms, and inter-agent communication
Prompt Engineering & LLM Configuration:
Writing sophisticated instructions (chain-of-thought, CoT) for each agent
Fine-tuning parameters (temperature, top-p, max_tokens) based on context
Iterative testing to optimize response quality
Cloud Integration Phase
OVH Cloud AI API: Setting up secure connection to proprietary endpoints
Data Pipelines: Connecting to existing business data (databases, data warehouses) to feed agents with context
Infrastructure: Configuration for scalability and observability
Key Deliverables
Detailed architecture document with orchestration diagrams
Functional and tested multi-agent system
Optimized prompt library (documentation + use cases)
Reproducible cloud configuration + operational guide
Performance metrics and improvement recommendations
Prompt engineering IT Architecture AI Agent Technical project management Python

Check out Hichem's experience

Xavier

Nodac

Reviewed on 7/15/2025

Excellent freelancer!! A mastery of his subject, and expertise that he enjoys sharing on ML, which I really needed. The work is clean and communication is great. Deadlines slipped a bit, but that's an inherent problem with tech solutions, involving back-and-forth. You have to be flexible!! I will work with Hichem again whenever I can. Thank you very much for your services.