Yacine Mekideche

Creation and development of the AI ecosystem for YMUNIT, the first French platform for automated cybersecurity.

Integration of LLMs, advanced NLP processing, and RAG systems on a hybrid architecture (on-premise & SaaS) to automate risk management, compliance (EBIOS RM, ISO 27001), and audit processes.

This approach significantly reduced risk exposure times, accelerated EBIOS analyses (×18), generated +€250k/year in savings for clients, and brought sovereign AI components (internal LLM, secure RAG) into production, while strengthening R&D.

• Structuring MITRE cyber databases via scraping and ingestion (BeautifulSoup, Pandas, SQLAlchemy)

• AI automation of the platform: NER, summarization, classification, Q&A (OpenAI, LangChain, Prompt Engineering, FastAPI)

• Automatic generation of PDF reports (Jupyter, ReportLab, FPDF, WeasyPrint)

• Hybrid RAG deployment: MongoDB, FAISS, LangChain, Streamlit, AWS (Lambda, S3, Bedrock, API Gateway)

• Development of a secure internal LLM (HuggingFace, PyTorch, TensorFlow – deployment via SageMaker, EC2)

• Design of 24/7 RAG AI agents (n8n, ElevenLabs API, SpeechRecognition, Amazon Polly)

Predictive analysis of criminal trends using the CEZAR database, as part of the Paris 2024 Olympic Games, to enhance railway security.

Implementation of an automated sensitive zone detection tool, which helped prioritize interventions and reduce operational costs.

• Development of a Streamlit application for extraction, processing, and automated reporting of CEZAR data (Python, Pandas, Plotly, Matplotlib)

• Supervised and unsupervised ML modeling to identify high-risk stations (XGBoost, K-means, Scikit-learn – 75% accuracy)

• Creation of a real-time dashboard for monitoring Emergency Stop Buttons with multi-source integration (SQL, Web Scraping, DAX, Power BI)

Predictive analysis of the lifespan of frozen soap bubbles, carried out as part of an experimental scientific study.

Implementation of a modular, replicable, and interpretable pipeline to identify key features of their degradation.

• Data processing: cleaning, normalization (Python, Pandas, NumPy)

• Advanced statistical analysis: correlations, anomalies, tests (scipy.stats, z-score, boxplots)

• Supervised predictive models: Random Forest, Gradient Boosting (80% accuracy)

• Data Visualization & reporting (Matplotlib, Seaborn, Power BI)