Freelancer profile translated to English.
Back to original languageAbout Christian
AI & MLOps Engineer | 10 years of experience | LLM, RAG, Fine-tuning in production
An AI POC that works in demo is not a production product at scale. Expert in designing and industrializing AI solutions for several million users, I combine 10 years of technical experience with sharp expertise in MLOps, GenAI, and Data Engineering.
Why choose me?
Proven production: AI solutions deployed for several million users with 24/7 monitoring and incident management
Modern AI Stack: LLM (OpenAI, LLaMA), RAG, NLP, Computer Vision, Whisper
MLOps Approach: Complete pipelines from development to production with CI/CD, versioning (MLflow), and automated retraining
Guaranteed Reliability: BDD methodology (Behave, PyTestBDD) for impeccable quality
Key Achievements
Bi-modal intelligent chatbotwith RAG architecture (OpenAI + pgvector) and Gradio interface
Transcription and generation APIs(WhisperCPP, LLamaCPP) in local container
Complete Interactive Voice Response systemhandling millions of calls in production
Critical ETL pipelineshigh volume with strong performance constraints
Complete MLOps infrastructure: monitoring, alerting, operating procedures
Technical Stack
AI & ML: Python, PyTorch, Scikit-learn, Hugging Face, LangChain, Rasa, azureOpenAI ...
Data: Pandas, SQL, pgvector, Kafka, ETL ...
DevOps/MLOps: Docker, OpenShift, FastAPI, Ansible, MLflow, Behave, PyTestBDD ...
Cloud: AWS, Azure, Supabase
My Approach
I don't just develop models - I industrialize them to work reliably and performantly in production. Close collaboration with business teams, rigorous documentation, and maintainable code are at the heart of my approach.
Available for consulting, development, and industrialization missions for your Data/AI projects.
French
Native or bilingual
English
Fluent
Can work on-site
Paris (up to 50km), Niort (up to 50km), Bordeaux (up to 50km), Nantes (up to 50km), Toulouse (up to 50km)
Experience
- steerway
On Malt
Creation of an LLM fine-tuning/evaluation pipeline for a content completion assistantSOFTWARE PUBLISHINGMarch 2026 - April 2026 (1 month)Dijon, FranceFrom scratch creation of a complete fine-tuning and evaluation pipeline for open-source LLM models (4B to 27B parameters) for a structured content completion assistant.✅ End-to-end pipeline- Config-driven Python package (writable YAML) with CLI: train → merge → eval → analysis, orchestrated by automated script chain- 3 inference backends implemented (SGLang, vLLM, HuggingFace) with cross-framework incompatibility management- MLflow Model Registry: versioned LoRA adapters, linked training + eval metrics, complete traceability- Detailed evaluation metrics: semantic similarity, textual overlap, perplexity, latency, throughput, length calibration- Reproducible exploratory data analysis (EDA) module✅ Systematic benchmarking of 14 models- 14 models evaluated in zero-shot, 10 fine-tuned, 30+ configurations tested- Methodical exploration: datasets of increasing sizes, 1-5 epochs, LoRA/DoRA, ranks 8-32, learning rates, dataset rebalancing- Measured fine-tuning gain: +15 points in semantic similarity- Report with model recommendations and quality/latency/cost trade-offs✅ Debugging and technical investigations- Root cause analysis on 10+ complex issues (framework incompatibilities, OOM, catastrophic forgetting)- Comparative benchmark of training frameworks (Unsloth vs Axolotl vs TorchTune)- Deployment and management of runs on Scaleway GPUs (L4 24GB, L40S 48GB, H100 80GB): VRAM tuning per model, auto-shutdown✅ Delivery and process- Complete and autonomous deliverables, validated with the client at each step- Tested code (unit + integration), complete technical documentation- Mission report with recommendations and model ranking- Integration into the client's existing workflow and toolsStack: Python, PyTorch, Unsloth, LoRA/DoRA, SGLang, vLLM, MLflow, MongoDB, Scaleway GPU - MAIFData AI EngineerBANKING AND INSURANCEApril 2023 - Today (3 years and 2 months)Niort, France
Data & Artificial Intelligence Engineer
Main missions
Design, development, and industrialization of AI solutions in production for several million users, with a comprehensive MLOps approach.Key achievements
Large-scale Interactive Voice Response (IVR) system — Complete Rewrite (2026)Migration to a full custom Python solution, in production for several million users. Event-driven architecture (Kafka), conversational tree exposed as an API, internal classifier, Redis persistence. Inspired by Cosmic Python: hexagonal architecture, domain/services/adapters separation, unit of work, repository pattern. Result: improved maintainability, business/infra decoupling, native testability.Bi-modal intelligent chatbotChatbot with traditional and RAG (Retrieval Augmented Generation) modes, based on OpenAI, pgvector, and Python. Smooth and intuitive Gradio interface.High-performance AI APIsTranscription (WhisperCPP) and text generation (LLamaCPP) APIs with FastAPI, deployed on OpenShift via optimized Docker images. Hosted locally for privacy and performance.MLOps infrastructureComplete MLOps pipelines: monitoring, operating procedures, incident management, model retraining. MLflow for versioning and performance tracking.Continuous Improvement ToolsSampling and annotation tools to optimize model training. Vector database integrated with Kafka for data exploration and analysis.Business AnalysesRegular statistical studies to inform decisions and prioritize user needs.Technical StackPython, Azure OpenAI, LLaMA, Whisper, FastAPI, Rasa, pgvector, Kafka, Redis, Docker, OpenShift, MLflow, Behave, PyTestBDD, Gradio, Pandas, PyTorch, Scikit-learn, Hugging Face, LangChain, NLP, Hexagonal architecture, Event-driven, Legacy refactoring - Kelguichet.comInteractive visualization module for a Flutter mobile application in productionSOFTWARE PUBLISHINGNovember 2025 - November 2025Lyon, France✅Complete module development(13 incremental phases)- 2 distinct visualization modes (rich list view + interactive graph view)- Cascade navigation with optimized state management- Dual filtering/prioritization system- Real-time search with automatic scroll to results- 3-mode demo system (API + 2 test datasets)✅Quality Architecture & Code- Creation of dedicated helpers to eliminate 100% of code duplication- Complete refactoring for 2x maintainability- Reusable and testable patterns✅Optimization & Performance- CustomPainter for optimized graphic rendering- Smooth 2D scrolling tested with a volumetric dataset (300+ elements)- Elegant handling of edge cases (missing data, circular structures)✅Process & Communication- Regular client validationsResult: Module delivered, validated, and deployed
Reviews
Recommendations
Be the first to recommend Christian
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Doctorate in Theoretical and Phenomenological PhysicsUniversity of Caen2011
- Master in Data ScienceCentraleSupélec/OpenclassRoom2023📊 Formation Data Scientist Diplôme de niveau Master (Bac+5) certifié RNCP En partenariat avec CentraleSupélec 🎯 Parcours intensif 600 heures de formation (9 mois à temps plein) Formation professionnalisante couvrant l'ensemble de l'écosystème Data Science 💻 Compétences techniques maîtrisées Programmation & Analyse Python avancé pour la Data Science (NumPy, Pandas, Matplotlib, Scikit-learn) Statistiques descriptives et inférentielles Manipulation de données structurées et non structurées Machine Learning Apprentissage supervisé : régression linéaire/logistique, SVM, arbres de décision Apprentissage non supervisé : clustering (K-means, DBSCAN), réduction dimensionnelle (ACP, t-SNE) Méthodes ensemblistes : Random Forest, Gradient Boosting, XGBoost Deep Learning Réseaux de neurones et perceptrons Computer Vision : classification d'images (CNN) NLP : traitement du texte (Bag of Words, Word Embedding) Big Data & Déploiement Cloud Computing : AWS (Amazon Web Services) Calcul distribué : PySpark, Hadoop MapReduce MLOps : déploiement via API REST, dashboards interactifs Versioning : Git & GitHub 🚀 Projets professionnels réalisés ✅ Analyses statistiques multivariées pour recommandations stratégiques ✅ Modèles prédictifs (consommation énergétique, scoring crédit) ✅ Segmentation client pour optimisation marketing ✅ Classification automatique de biens de consommation (images + texte) ✅ Architectures Big Data pour traitement de données massives ✅ Mise en production de modèles ML via API et dashboards 🎓 Compétences transversales 📈 Data Visualization et storytelling 📝 Communication de résultats techniques à des non-experts 🔄 Gestion de projets data de bout en bout 🤝 Méthodologies agiles Cette formation m'a permis d'acquérir une expertise complète du métier de Data Scientist, de l'exploration des données à la mise en production d'algorithmes d'IA.