Results-oriented AI Engineer turning complex data into production-ready solutions.
Based in France 🇫🇷
Currently working on : Veona, an open-source monitoring platform built with Go and TypeScript, Selma, a React + D3 template for visualising taxonomic and DAG trees, and ChartAtlas, a Rosetta Stone-style chart library:
AI engineer and data scientist focused on solving complex business through the design and industrialization of ML/AI solutions. I combine technical expertise in AI/machine learning, MLOps and software developement with a rigorous approach to bridge the gap between experimental innovation and reliable production systems.
- Core domains: Machine Learning, GenAI, embedding models, audio processing.
- Engineering: End-to-end ML pipelines — from data collection to model deployment (local or Cloud LLM serving).
- I value mentorship & knowledge transfer, creating technical documentation as well as educational materials to demystify AI.
I work with a wide range of tools for data science, engineering, and MLOps.
Other key skills: LLMs (RAG, Embeddings), NLP (SpaCy, NLTK), Vector DBs (ChromaDB, FAISS), Airflow, ELK Stack.
Here are some projects I have engineered or contributed to, from ideation to production.
(click to expand)
Veona — Push-based ML powered system monitoring platform
Two main components:
- A Go agent that collects system metrics — CPU, memory, disk, network, GPU, processes, battery, entropy, NTP drift — and ships them via compressed HTTP batches.
- A TypeScript/Hono server that validates authentication, transforms metrics to Prometheus format, and forwards them to VictoriaMetrics for time-series storage.
What's next: Machine learning capabilities are planned for a second phase — anomaly detection on metric streams, predictive alerting using trained models, and automated baseline profiling of system behavior across fleets of agents.
ASR.lab — Multi-engine ASR benchmarking framework
Multi-engine ASR benchmarking framework with configurable noise, enhancement and normalization conditions. Compares Whisper, Wav2Vec2, NeMo, and other backends on custom datasets — outputs multiple metrics, charts, and vizualizations in an interactive HTML report for model selection.
StellaScript — Fully local audio transcription pipeline
Fully local audio transcription pipeline for sensitive data. Source separation, speaker diarization, timestamped transcription (with three granularities) — runs entirely on-premise.
flac_toolkit — CLI for FLAC audio file management
CLI tool for bulk FLAC file management: integrity analysis, repair, ReplayGain tagging, and content-based duplicate detection via audio fingerprinting. Designed for large-scale library maintenance.
ForzaEmbed — Embedding benchmark tool for RAG pipelines
Grid-search evaluation tool for RAG embedding selection. Benchmarks multiple embedding models on custom corpora, evaluating retrieval quality via metrics and interactive textual heatmap.
SmartWatch — Automated web scraping + LLM extraction pipeline
Automated pipeline combining web scraping with LLM-based structured extraction. Reduces manual data processing by up to 75% — transforms unstructured HTML into validated JSON output.
PRISM — Pattern Recognition & Intelligent Similarity Matcher
Composable Python library for string similarity matching. Supports edit distance (Levenshtein, Damerau-Levenshtein, Hamming), sequence similarity (Jaro-Winkler), token-based (TF-IDF), phonetic (Soundex, Metaphone) and semantic similarity (Jina Embeddings) with unified API.
Friands — Restaurant analytics and recommendation engine (Contributor)
Restaurant discovery and analytics app combining NLP-powered sentiment analysis with geographic visualization. Features automated review summarization, multi-criteria filtering, and interactive dashboards — leveraging Mistral AI and fine-tuned BERT for insights.
SmartRescue — AI assistant for emergency call centers (Contributor)
SmartRescue is an AI-powered assistant for emergency call centers. It combines live speech-to-text with RAG to provide real-time medical guidance and support operators during critical calls.
Selma — Hierarchical taxonomy and DAG visualizer TypeScript template.
Interactive explorer template for hierarchical and directed acyclic graph (DAG) taxonomies. Integrates D3.js visualizations with Markdown-driven content to navigate complex relationships and present item details. Includes multilingual support, a searchable sidebar, breadcrumb navigation, and a responsive Markdown viewer.
ChartAtlas — Interactive chart catalogue; source code for each chart type.
Inspired by the Financial Times' "visual vocabulary", I've compiled these charts into a dynamic interface and included code from five different programming libraries to generate them. This tool is designed for data analysts to select and create their own charts, and it serves as a Rosetta Stone for Python (Matplotlib and Seaborn), R, D3.js, and Julia.
School of statistics — Interactively explore statistical and machine learning concepts
Interactive visualizations to intuitively grasp statistical and machine learning concepts. Real-time parameter tuning with live updates.
MIMOSA — Logistic regression R package with built-in preprocessing (Contributor)
R package implementing logistic regression from scratch, with integrated preprocessing pipeline: missing value imputation, categorical encoding, and normalization — no external preprocessing required.
ShadowLog — AI-powered log analysis platform (Contributor)
Log analysis platform using AI-powered clustering and statistical anomaly detection to identify anomalies and patterns in large-scale log files. Includes temporal analysis, keyword-based error detection, and consecutive error sequence identification.
ORBIT — Organized Repositories Based on Images Timing
Organize photos into a structured directory tree based on their EXIF metadata.
eml2pdf — Compact multiple emails into a single pdf
Convert and compact multiple .eml files into a single pdf document.
- M.S. in Data Science (Distinction) - Lumière University Lyon 2 (2025)
- M.S. in Biomedical Technology (Distinction) - University of Technology of Compiègne (2003)
- M.S. in Applied Physics - Grenoble Alpes University (2002)