GenAI & LLMs
RAG assistants and agentic workflows in production — retrieval design, prompt and context engineering, evaluation harnesses, and reverse-explainability for LLM answers.
LangChainLangGraphllama-indexpgvectorVertex AI
Machine Learning Engineer · Paris, France
Production AI for
luxury Maisons.
I'm Mohamed Niang — ML Engineer at LVMH, where I design and ship generative AI and recommendation systems that reach the Group's Maisons, and founder of SetAIComply, making the EU AI Act workable for European SMEs.
Currently LVMH — ATOM AI platform
Previously IBM · Carrefour · FeetMe
Trained at Université Paris-Saclay · ENSAE
6+
Years in ML & data
60+
Open-source repos
19
Certifications
2
Products founded
01 About
From research notebook
to production system.
Mohamed Niang
ML Engineer @ LVMH · Founder @ SetAIComply
I build machine learning systems that actually reach users. At LVMH I work on the ATOM AI platform, turning research into products the Group's Maisons rely on every day — clienteling assistants, recommendation engines, forecasting, marketing science, computer vision and NLP — shipped as reusable Dataiku DSS plugins and Google Cloud micro-services.
My path runs from statistics to engineering: a statistical engineering degree at ENSAE Pierre Ndiaye, a Master's in Data Science at Université Paris-Saclay, then data science at FeetMe and Carrefour, and MLOps for recommender systems at IBM. That mix is why I care as much about evaluation, monitoring and documentation as about the model itself.
In 2026 I founded SetAIComply, an AI-Act-native compliance platform for European SMEs. Working on it convinced me that Responsible AI is an engineering problem before it is a legal one — which is exactly where I like to work.
- Based in
- Paris, Île-de-France · Hybrid & remote
- Role
- ML Engineer @ LVMH · Founder @ SetAIComply
- Focus
- GenAI & LLMs · Recommender systems · MLOps · AI governance
- Languages
- French · English
- Writing
- Medium
02 Expertise
What I do
Six areas where I have shipped systems into production, not just prototypes.
Recommender systems
Two-Tower retrieval with Deep & Cross Networks for cold-start, sequential models tuned with Direct Preference Optimization, embedding-based companion products and style clustering.
TF RecommendersDCNLSTM + DPOWord2Vec
Computer vision & NLP
Product image embeddings with CLIP and NASNet, colour quantization and blur scoring with OpenCV, multilingual text embeddings, topic and keyword extraction at catalogue scale.
CLIPOpenCVsentence-transformersBERTopicUMAP
Marketing science
Deep-learning attribution models and Marketing Mix Modeling in both Bayesian and OLS flavours, connecting media investment to measurable commercial outcomes.
MMMBayesian inferenceAttributionUplift
MLOps & platform
Reusable plugins and micro-services, experiment tracking, hyper-parameter search, containerised deployments, CI/CD and observability — so a model keeps working after launch day.
Dataiku DSSMLflowDockerFastAPIGitHub ActionsOpenTelemetry
AI governance
EU AI Act conformity in practice: risk classification, Annex IV technical documentation, evidence automation and Responsible AI guardrails designed for teams without a legal department.
EU AI ActAnnex IVRisk classificationModel cards
03 Experience
Where I have worked
Machine Learning Engineer
LVMH- ML Engineer on the ATOM AI platform, delivering production ML & GenAI products to the Group's Maisons as reusable Dataiku DSS plugins and GCP micro-services.
- Built a RAG-powered clienteling assistant for advisors (LangChain, LangGraph, Vertex AI, FastAPI, PostgreSQL/pgvector) and a Responsible-AI reverse-explainability layer for LLM answers.
- Designed Two-Tower retrieval with Deep & Cross Networks for cold-start, an LSTM sequential recommender trained with Direct Preference Optimization, and embedding-based companion products.
- Shipped computer vision and NLP pipelines: CLIP and NASNet image embeddings, colour quantization and blur scoring, multilingual embeddings, BERTopic and KeyBERT.
- Delivered marketing science models: deep-learning attribution and Bayesian / OLS Marketing Mix Modeling.
Founder
SetAIComply- Founded an AI-Act-native compliance platform for European SMEs: classify AI systems, generate Annex IV technical documentation and stay compliant in 24 EU languages.
- Self-serve product built on FastAPI and Next.js, with a compliance lab producing continuous CI evidence for a governed AI system.
Senior Machine Learning Engineer
IBM- Built and maintained AI software modules for client projects (SharePoint and Azure I/O connectors, Snowflake data loaders) with unit tests and Sphinx documentation.
- Contributed to a ChatGPT-style generative AI application: Cosmos DB backend for chats and users, plus UX work on the web app.
- Led data and module migrations (Azure SQL → Snowflake, AzureRM → Az), production bug fixing and code-quality improvements.
Machine Learning Engineer
IBM- Pure MLOps for recommender systems: "customers also bought", complementary products, similar products and product search.
- Engineered containerisation, CI/CD pipelines, runbooks, data loaders, tracing and web services on Microsoft Azure (Cosmos DB, App Services, Batch, Application Insights).
Data Scientist
Carrefour- Built fraud-prevention analytics on Google Cloud: advanced BigQuery models, automated ETL pipelines triggered through Pub/Sub and Cloud Functions, reporting dashboards.
- Automated email fraud alerts with n8n and web scraping; applied unsupervised anomaly detection for labelling and supervised deep learning for prediction.
- Operated the full Vertex AI lifecycle: custom training jobs, TensorBoard comparison, endpoints, monitoring and pipelines.
Data Scientist
FeetMe- Developed a feature-engineering module for human activity recognition from smartphone and wearable sensor recordings.
- Modelled activity with penalized elastic-net logistic regression and decision trees under a strong validation protocol, on a GCP data stack.
04 Selected work
Products & platforms
Things I have built or am building — in production, in the open, or in progress.
Founder · 2026
SetAI
Comply
EU AI Act · SaaS
Live
SetAIComply
AI-Act-native compliance software for European SMEs. Classify your AI systems against the risk tiers, generate Annex IV technical documentation, and keep evidence current — self-serve, in 24 EU languages, starting from €0. Built with FastAPI and Next.js, with a compliance lab that produces continuous CI evidence.
FastAPINext.jsTypeScriptPostgreSQLEU AI Act
Visit setaicomply.com
LVMH · ATOM AI platform
Clienteling
& recommendation
GenAI · RecSys · Responsible AI
Ongoing
Production AI for the Maisons
A family of ML and GenAI products delivered on LVMH's ATOM AI platform: a RAG clienteling assistant for advisors, Two-Tower and sequential recommenders, product image and text embeddings, and a reverse-explainability layer that makes LLM answers auditable. Packaged as reusable Dataiku DSS plugins and GCP micro-services so several Maisons can adopt the same building blocks.
LangGraphVertex AITensorFlow RecommenderspgvectorDataiku DSS
Internal platform — no confidential detail shared.
Side project · 2026
Citez
Moi
AI visibility · SEO for LLMs
In development
CitezMoi
Get cited by ChatGPT. An AI-visibility audit for French-speaking SMEs and agencies: measure how — and whether — assistants mention your brand, receive generated fixes, and track the effect week after week.
LLM evaluationWeb crawlingReporting
05 Open source
Code in the open
A selection from 60+ public repositories — mostly hands-on notebooks I wrote while learning something properly.
Computer Vision for Autonomous Cars
How self-driving perception actually works: lane detection, object detection and tracking, LiDAR — explained and implemented.
Reinforcement Learning Specialization
A complete walk-through of the University of Alberta Fundamentals of Reinforcement Learning specialization.
GCP Professional Data Engineer Path
The full learning path I followed to become Google Cloud Professional Data Engineer certified, with labs and notes.
Human Activity Recognition
Activity recognition from wearable inertial sensor networks — feature engineering, model selection and a rigorous validation protocol.
Virtual Adversarial Training
A Keras implementation of the original VAT paper for semi-supervised learning, reproduced from scratch.
Data Science With Python
A practical introduction to data science with Python: pandas, NumPy and matplotlib, from first import to first model.
06 Writing
Notes & tutorials
I write on Medium about the tools I use day to day — mostly the explanation I wish I had found first.
07 Education
Training & credentials
2019 — 2020
MSc Mathematics & Applications — Data Science
Université Paris-Saclay
Co-accredited with CentraleSupélec, École Polytechnique, ENS Paris-Saclay, ENSAE Paris, Télécom SudParis and ENSIIE. Machine learning, deep learning, Bayesian learning, high-dimensional statistics, optimization, GPU programming and big data.
2015 — 2019
Engineering degree — Statistics & Business Analytics
ENSAE Pierre Ndiaye
Probability, estimation and test theory, econometrics, time series, survey theory, data mining, databases and data warehousing, statistical software (R, SAS, Stata).
19 Certifications
- Professional Data Engineer — Google Cloud
- Machine Learning — Duke University
- NLP with Classification and Vector Spaces — DeepLearning.AI
- How Google Does Machine Learning — Google
- Smart Analytics, ML and AI on Google Cloud — Google
- Serverless Data Processing with Dataflow (3 parts) — Google
- Modernizing Data Lakes and Data Warehouses — Google
- Building Resilient Streaming Analytics Systems — Google
- Enterprise Design Thinking — Co-Creator & Practitioner — IBM
- Exploratory Multivariate Data Analysis — MyMoocs
08 Contact
Let's build something
worth shipping.
I'm always happy to talk about applied GenAI in production, recommender systems at scale, or making the EU AI Act workable for small teams. Speaking, collaboration and interesting problems all welcome.