I industrialize your AI systems
scoping → POC → production
AI & ML Engineer (3.5 years of XP) and 2× AWS Certified, I build and run your AI solutions, on the cloud and on-prem. Concrete, sovereign, high-performing.
Live from my desk
Chat with my 3D avatar, check how your needs match my profile, or upload a document for analysis.
For complex or confidential requests, contact me directly.
Data & memory
Talk with my AI avatar, live
Your image here
Your AI POC is ready?
INITIALIZING PIPELINE...
0%
PRODUCTION LIVE
0:00
Your voice is never recorded: only the transcribed text is kept.
Connecting…
Oops, something went wrong. Try again in a moment.
My Services
Sovereign & On-Premise AI
Your models and data under your control, down to air-gapped environments.
GenAI & AI Agents
Tool-equipped assistants, agentic workflows and business process automation.
Architecture & AI Integration
Robust, modular systems, fully integrated into your information system.
RAG & Augmented Search
Reliable, sourced answers grounded in your own data.
AWS, MLOps & LLMOps
Automated, reproducible deployments, ready for production.
Evaluation, Security & Observability
Quality, risk and performance under control end to end.
Vision AI & Document AI
Documents, images and scans analyzed and turned into usable, structured information.
Fine-Tuning & Model Adaptation
Models specialised for your use cases and business constraints.
Machine Learning & Data
From raw data to predictive models and decision support.
Sovereign & On-Premise AI
Your models and data under your control, down to air-gapped environments.
GenAI & AI Agents
Tool-equipped assistants, agentic workflows and business process automation.
Architecture & AI Integration
Robust, modular systems, fully integrated into your information system.
RAG & Augmented Search
Reliable, sourced answers grounded in your own data.
AWS, MLOps & LLMOps
Automated, reproducible deployments, ready for production.
Evaluation, Security & Observability
Quality, risk and performance under control end to end.
Vision AI & Document AI
Documents, images and scans analyzed and turned into usable, structured information.
Fine-Tuning & Model Adaptation
Models specialised for your use cases and business constraints.
Machine Learning & Data
From raw data to predictive models and decision support.
From Machine Learning to LLMOps, in the field.
YMUNITB2B SaaS, GRC & cybersecurity
Technical lead for the AI components, from business scoping to industrialization on AWS and client infrastructure. Trade-offs between quality, performance, cost and confidentiality with the product, development and infrastructure teams.
- Local LLM serving : air-gapped deployment with vLLM on H100 80 GB; quantization, FP8 KV cache and prefix caching. Result : 1.8× throughput, -40% inference cost at constant latency, -45% VRAM footprint.
- Industrialization : environment automation with Terraform, Docker and GitLab CI/CD, together with the infrastructure team. Preparation of offline installations. Result : Provisioning cut to under an hour, three reproducible environments.
- Model selection : benchmark of around ten open and proprietary LLMs; prompt and parameter optimization per use case. Result : Cost per token -30% at equivalent evaluated quality.
- Evaluation and observability : load, concurrency and endurance tests; factuality checks, output validation, traceability and monitoring of errors and performance.
- Agentic : contribution to a LangGraph workflow tracking audit actions; evidence retrieval in client tools through tool calling and MCP, state management and proposed updates.
- Delivery and knowledge transfer : definition of AI interfaces, code reviews, task tracking and estimates, mentoring of a junior engineer; documentation of technical choices and operating procedures.
- Stack :
- Python · FastAPI · LangGraph · MCP · Mistral · vLLM · H100 · AWS · Docker · Terraform · GitLab CI/CD · CloudWatch
- SaaS / AWS :
- Bedrock · Titan Embeddings · Lambda · ECS/Fargate · EC2 · S3 · API Gateway · CloudWatch
- On-premise :
- Docling (parsing/OCR) · BGE-M3 (embeddings) · BGE-reranker · Mistral via vLLM · NVIDIA H100 GPU · Internal storage · Docker · Air-gapped installation
YMUNITB2B SaaS, GRC & cybersecurity
Development and integration of AI features for document analysis, compliance audits and action plans, from prototype to production.
- Audit automation : production of a first sourced document analysis, before business validation. Result : About 30 minutes, versus several weeks to several months of manual work depending on the scope.
- RAG in production : multi-format ingestion, OCR, chunking, metadata, embeddings, hybrid search, reranking and citations. Result : Recall@7 = 0.95 on the compliance analysis module.
- Backend and integration : development of Python/FastAPI services consumed by the application; API contracts, Pydantic schemas and validation of structured inputs and outputs.
- Prompt engineering : extraction, classification, summarization and Q&A; building the document context, business instructions and JSON generation for application processing.
- Cyber analysis : framework mapping with embeddings, LLM classification and consistency rules, with human validation; use of CVE and MITRE ATT&CK to support prioritization and recommendations.
- Quality and maintenance : evaluation of retrieval, factuality and citations; tests, merge requests, journey validation with developers, bug fixing and technical documentation.
- Stack :
- Python · FastAPI · Pydantic · LangChain · Docling · MongoDB · Titan Embeddings / Bedrock · BGE-M3 · BGE-reranker · S3 · Docker · GitLab
- SaaS / AWS :
- Bedrock · Titan Embeddings · Lambda · ECS/Fargate · EC2 · S3 · API Gateway · CloudWatch
- On-premise :
- Docling (parsing/OCR) · BGE-M3 (embeddings) · BGE-reranker · Mistral via vLLM · NVIDIA H100 GPU · Internal storage · Docker · Air-gapped installation
SNCF Voyageurs
Predictive analysis of railway security risks to support preparations for the Paris 2024 Olympic and Paralympic Games.
- Data preparation : collection, structuring and automation of processing to feed analyses and reporting.
- ML modeling : development of predictive models to identify at-risk stations and guide operational analysis priorities.
- Decision-support tools : creation of a Streamlit application and dashboards to explore results and track security indicators.
- Stack :
- Python · Pandas · Scikit-learn · XGBoost · K-means · SQL · BeautifulSoup · Streamlit · Power BI · DAX · Plotly
CNRSMSC Laboratory, UMR 7057
Predictive modeling of the lifetime of frozen soap bubbles as part of a physics research project.
- Experimental data : structuring and preparation of observations for statistical analysis and modeling.
- Machine Learning : development and comparison of supervised models, notably Random Forest and Gradient Boosting, to estimate bubble lifetime.
- Results analysis : evaluation of predictions and study of the models' limits on experimental data.
- Stack :
- Python · Pandas · NumPy · SciPy · Scikit-learn · Random Forest · Gradient Boosting