My Projects

Explore my portfolio of Secure On-Premise AI solutions, MLOps implementations, and robust software architectures.

🚀

NotationAI — Industrial AI Scoring System

Architected a production-grade, two-stage AI pipeline for automated evaluation processing in an air-gapped intranet. Features a fairness filter using LLMs for semantic refactoring and a high-performance quantized CamemBERTav2 engine for real-time inference on CPU.

PythonCamemBERTav2ONNX RuntimeFastAPIReactDockerDVC
🚀

Secure On-Premise LLM Infrastructure

Integrated Large Language Models into a private intranet environment using local inference engines (Ollama) and n8n orchestration. Built secure RAG pipelines with PostgreSQL VectorDB for document summarization and judicial assistance without cloud exposure.

Ollaman8nPostgreSQLVectorDBDockerPrompt Engineering
🚀

MIW — Bio-Chaotic Medical Steganography

Developed a VAE-GAN-based medical image watermarking system for PhD research. Implemented bio-chaotic encryption for secure payload embedding, achieving high PSNR (>51dB) and low BER on industrial GPU hardening runs (RTX 5090).

PyTorchVAE-GANONNXHydraDockerMedical Imaging
🚀

P.P.A — Training Management Platform

Full-stack platform for HACCP/Qualiopi-compliant training organizations in France. Features automated registration flows, funding management, QCM scoring, and email notifications. Built with Next.js 15+ and PostgreSQL.

Next.jsTypeScriptDrizzle ORMPostgreSQLTailwind CSSDocker
🚀

GestAut — Real-Time Gesture Recognition

High-performance gesture detection platform using YOLO11 and YOLOv8 models. Built with a FastAPI microservices backend for GPU inference and a React frontend for real-time webcam interaction.

PythonYOLO11YOLOv8FastAPIReactPyTorchCUDA
🚀

NLP MLOps Spam Detection System

End-to-end MLOps pipeline for French text classification. Includes DVC-tracked data versioning, ONNX optimization for CPU inference, and Prometheus-based monitoring for model drift detection.

PythonCamemBERTav2DVCFastAPIKubernetesPrometheus
🚀

National Administrative Platform

Led the development of a national platform for managing administrative requests using a microservices architecture. Implemented secure RBAC systems, Node.js/Express backends, and Alfresco API integration for document management.

Node.jsExpressSQL ServerAlfresco APIMicroservicesRBAC
🚀

Personnel Assignment Optimization (MIP)

Designed and implemented automated personnel assignment models using Mixed-Integer Programming (MIP). Solved complex combinatorial optimization problems with PySCIPOpt, significantly reducing manual scheduling overhead.

PythonPySCIPOptOptimizationMIPPandasTkinter
🚀

GPU-Accelerated Face Recognition

Face detection and recognition system using InsightFace (ResNet100). Features real-time inference on ONNX Runtime GPU and a desktop client for person registration with 512-dim embeddings.

InsightFaceONNX RuntimeCUDAFastAPITkinterSQLite
🚀

Bodycame — Secure VMS with Geolocation

Video Management System for body camera footage featuring homography-based geolocation mapping. Implements AES-256 encryption, RBAC, and secure audit logging for standalone Windows deployment.

PythonOpenCVTkinterSQLitePydanticAES-256
🚀

Explainable AI Bank Fraud Detection

XAI-focused fraud detection system using a 7-dimensional feature engineering pipeline. Achieved high Precision (0.90) while ensuring full model traceability for financial regulatory compliance.

Pythonscikit-learnExplainable AIPandasJupyter
🚀

Sorec — Equine Image Analysis Dashboard

Computer vision pipeline for horse marking identification. Features HSV color segmentation, morphological cleanup, and an automated 2x2 analysis dashboard for veterinary reporting.

PythonOpenCVNumPyComputer VisionMatplotlib