AWS Certified Machine Learning – Associate

William Roux

Machine Learning & MLOps Engineer Ingénieur Machine Learning & MLOps

Passionate about mathematics and computer science. Specialized in scalable machine learning systems, autonomous LLM agents, and robust MLOps pipelines. AWS Certified Machine Learning – Associate. Passionné par les mathématiques et l'informatique. Spécialisé dans la conception de systèmes ML scalables, d'agents LLM autonomes et de pipelines MLOps robustes. Certifié AWS Machine Learning – Associate.

William Roux

Work Experience Expérience Professionnelle

Apprentice ML & Software Engineer Apprenti ML & Software Engineer

Capgemini

2.5 years 2,5 ans
<>
Project SNCF Projet SNCF
1 year 1 an

Designed and built an AI assistant (RAG Chatbot) to streamline engine failure analysis, search complex technical documentation, and optimize maintenance schedules. Developed full-stack features using Java (Spring Boot) and Angular, and managed CI/CD pipelines via Jenkins. Conception et développement d'un assistant IA (Chatbot RAG) facilitant l'analyse des pannes, l'exploitation de la documentation technique et l'adaptation des schémas de maintenance des engins. Développement de fonctionnalités full-stack avec Java (Spring Boot) et Angular, et configuration de pipelines CI/CD sous Jenkins.

Java Spring Boot Angular RAG Jenkins Git
<>
Project ENEDIS Projet ENEDIS
1 year 1 an

Handled production MCO (maintenance) and resolved application anomalies on a proprietary PHP framework. Created and optimized complex SQL scripts for automated data processing and production fixes. Gestion du MCO (Maintien en Conditions Opérationnelles) en production. Résolution d'anomalies applicatives et développement de correctifs sur un framework PHP propriétaire. Création et optimisation de scripts SQL complexes pour le traitement automatique des données et les correctifs de production.

PHP SQL (Oracle) Bash Git
<>
Technical Upskilling Intermission & Formation
6 months 6 mois

Dedicated period for technical upskilling across multiple development environments. Successfully prepared for and obtained the AWS Certified Cloud Practitioner credential. Période d'intermission dédiée à la montée en compétences techniques sur différentes stacks et à l'obtention de la certification AWS Certified Cloud Practitioner.

Python AWS (Cloud Practitioner) Java Angular PHP

Featured Projects Projets Phares

AUTONOMOUS AI AGENTS & LLMs AGENTS IA AUTONOMES & LLM

Siquelator

Personal Perso

Autonomous local AI agent translating natural language to SQL with a self-debugging feedback loop. Agent IA autonome local traduisant le langage naturel en SQL avec boucle d'auto-correction. Development of an autonomous AI agent capable of translating natural language queries into complex SQL. The system integrates a background feedback loop that tests queries directly on the target database and deploys a self-correction pipeline (Self-Debugging) by analyzing error logs to iteratively reformulate its code. Designed to operate 100% locally and sovereignly. Développement d'un agent IA autonome capable de traduire des requêtes formulées en langage naturel en requêtes SQL complexes. Le système intègre une boucle de rétroaction s'exécutant en arrière-plan : il teste directement les requêtes sur la base de données cible et déploie un pipeline d'auto-correction (Self-Correction Loop) en analysant les traces d'erreurs générées pour reformuler son code de manière itérative. Conçu pour fonctionner de manière 100% locale et souveraine.

LangChain LLM Locaux Text-to-SQL Self-Debugging Gestion d'Agents
ADVANCED RAG & MLOPS RAG AVANCÉ & MLOPS

Astro-Retrieve

Personal Perso Private

Industrial RAG engine with fine-tuned embeddings, hybrid search, and FastAPI deployment. Moteur RAG industriel avec embeddings fine-tunés, recherche hybride et API FastAPI. Industrial-grade Retrieval-Augmented Generation (RAG) engine integrating a comprehensive MLOps workflow. Features custom fine-tuning of embedding models (PyTorch) and a hybrid search strategy (dense vector and lexical) connected to a local vector store, interfaced with LLMs via Ollama and FastAPI. Moteur de recherche et de génération augmentée (RAG) de niveau industriel intégrant une approche MLOps complète. Le projet inclut le fine-tuning personnalisé de modèles d'embedding (sous PyTorch) et s'appuie sur une stratégie de recherche hybride (dense/vectorielle et lexicale) connectée à une base vectorielle locale, interfacée avec des LLM via Ollama et FastAPI.

PyTorch Embedding Fine-Tuning Base vectorielle locale Ollama FastAPI Pipelines MLOps
NUTRITION & OPERATIONS RESEARCH NUTRITION & RECHERCHE OPÉRATIONNELLE

WePlate

Personal Perso Private

Intelligent meal recommendation hub generating custom dietary plans under strict nutritional constraints. Générateur intelligent de plans alimentaires sur-mesure sous contraintes macro-nutritionnelles. Intelligent meal recommendation and hydration tracking hub, optimized for individual or couple profiles. Generates tailored dietary plans by combining user preferences with strict macro-nutritional constraints indexed on physical and athletic goals. Hub intelligent de recommandation de repas et de suivi d'hydratation, optimisé pour des profils individuels ou de couples. Le système génère des plans alimentaires sur-mesure en combinant les préférences culinaires de l'utilisateur avec des contraintes macro-nutritionnelles strictes indexées sur ses objectifs physiques et sportifs.

PyTorch Deep Learning Recherche Opérationnelle (OR) Optimisation sous contraintes
NLP / ANOMALY DETECTION NLP / DÉTECTION D'ANOMALIES

Tune-Botrapoint

Personal Perso Private

NLP and Machine Learning pipeline auditing YouTube comments to detect bot campaigns and astroturfing. Pipeline NLP et ML détectant les campagnes d'astroturfing et de bots sur YouTube. End-to-end data analysis pipeline designed to scan and audit comments from the YouTube API to identify inauthentic influence campaigns (positive astroturfing) and compute a global bot score. Architecture segmented into four modules: Collector (YouTube API v3 ingestion), Behavior Analyzer (heuristics), Language Engine (NLP/embeddings), and Decision Maker (unsupervised ML). Pipeline complet d'analyse de données conçu pour scanner et auditer les commentaires issus de l'API YouTube afin d'identifier les campagnes d'influence inauthentiques (astroturfing positif) et d'attribuer un score de suspicion ("Score de Bot") global. Architecture segmentée en quatre modules : Collecteur (ingestion YouTube API v3), Analyseur de comportement (heuristiques de patterns), Moteur Linguistique (embeddings textuels) et Décideur (Machine Learning non-supervisé).

API YouTube Data v3 NLP Extraction de features Isolation Forest Scikit-Learn
GENETIC ALGORITHMS & MULTI-AGENT SYSTEMS ALGORITHMES GÉNÉTIQUES & SYSTÈMES MULTI-AGENTS

Strategenetique

Personal Perso

Multi-agent genetic algorithm simulating behavioral strategies and cooperation in the Prisoner's Dilemma. Simulation multi-agent d'un algorithme génétique appliqué au Dilemme du Prisonnier. Multi-agent simulation of a genetic algorithm applied to game theory, specifically the Iterated Prisoner's Dilemma. Built entirely from scratch without external frameworks, this project studies the evolution of strategies over generations and analyzes the conditions for the emergence and stabilization of cooperation. Simulation multi-agent d'un algorithme génétique appliquée à la théorie des jeux, et plus précisément au Dilemme du Prisonnier Itéré. Développé entièrement "from scratch" sans framework externe, ce projet étudie l'évolution des stratégies comportementales au fil des générations et analyse les conditions d'émergence et de stabilisation de la coopération face à des profils défecteurs.

Python NumPy (Vectorisation) Algorithmes génétiques Théorie des Jeux
COMPUTER VISION / REAL-TIME INFERENCE COMPUTER VISION / REAL-TIME INFERENCE

MoodX

Personal Perso

Real-time facial expression recognition pipeline classifying emotions from webcam video streams. Reconnaissance d'expressions faciales en temps réel sur flux webcam via CNN optimisé. Real-time Facial Expression Recognition (FER) pipeline operating on a live webcam video feed. Detects faces and instantly classifies 7 universal emotions using a Deep Learning model highly optimized for low-latency inference. Pipeline de reconnaissance des expressions faciales (FER - Facial Expression Recognition) opérant en temps réel sur un flux vidéo direct (webcam). Le système détecte les visages et classifie instantanément les 7 émotions universelles grâce à un modèle de Deep Learning hautement optimisé pour l'inférence à faible latence.

PyTorch OpenCV Real-time Inference Deep Learning (CNN)
COMPUTER VISION / MEDICAL IMAGING COMPUTER VISION / IMAGERIE MÉDICALE

Zoidberg 2.0

Epitech

Deep Learning medical imaging application detecting pneumonia indicators on chest X-rays. Analyse d'images médicales par Deep Learning pour la détection de la pneumonie sur radiographie. AI-driven medical image analysis application. The model processes chest X-rays to autonomously identify pathological indicators of pneumonia, demonstrating expertise in complex image classification and handling sensitive clinical datasets. Application d'analyse d'images médicales basée sur l'IA. Le modèle traite des radiographies pulmonaires afin d'identifier de manière autonome les indicateurs pathologiques liés à la pneumonie, validant des compétences en classification d'images complexes et en manipulation de datasets cliniques sensibles.

Computer Vision CNN Imagerie Médicale Deep Learning
LARGE LANGUAGE MODELS (LLMs) LARGE LANGUAGE MODELS (LLM)

Voltaire

Personal Perso

Autoregressive French Transformer model trained from scratch with Apple Silicon MLX optimization. Modèle de langage Transformer français autorégressif entraîné "from scratch" avec Apple MLX. Full training from scratch of a custom autoregressive French Language Model based on the Transformer architecture. Implements core attention mechanisms (Scaled Dot-Product Attention), causal masking, positional encoding, and native training optimization on Apple Silicon. Entraînement complet "from scratch" d'un modèle de langage autorégressif personnalisé pour le français de type Transformer. Ce projet d'ingénierie fondamentale a permis d'implémenter les briques de base de l'attention (Scaled Dot-Product Attention), les masques causaux, les encodages de position, ainsi que la gestion de l'entraînement matériel natif sur l'architecture Apple Silicon.

Apple MLX Framework Transformer custom Tokenisation native
AUDIO PROCESSING & BIOMETRICS AUDIO PROCESSING & BIOMÉTRIE

Diarizator

Personal Perso Private

Audio diarization and speaker recognition system using voice biometrics and embeddings. Système de diarisation audio et reconnaissance de locuteurs par signatures vocales. Speaker diarization system capable of isolating and segmenting different speakers in a continuous audio recording. Features a biometric enrollment module mapping names to unique voiceprints (Speaker Embeddings), enabling cross-recording speaker identification. Système de diarisation vocale capable d'isoler et de segmenter les différents locuteurs présents au sein d'un enregistrement audio continu. Intègre un module d'enrôlement biométrique permettant d'associer un nom à une signature ou empreinte vocale unique (Speaker Embedding), rendant possible l'identification automatique et transverse de ce locuteur dans n'importe quel flux audio futur.

Diarisation Audio Speaker Embeddings Clustering Adaptatif
NLP CHATBOT & GRAPHS CHATBOT NLP & GRAPHES

Travel Order Resolver

Epitech

Logistics conversational agent using NER for entity extraction and route optimization on graph networks. Agent conversationnel logistique utilisant le NER et le calcul d'itinéraires sur graphes. Logistics-oriented conversational agent. The chatbot classifies travel intent from queries, extracts named entities (departure and arrival cities) using a NER model, and computes optimal transport routes. Includes a custom graph network modeling option to operate without external APIs. Agent conversationnel orienté logistique. Le chatbot qualifie l'intention de déplacement à partir d'une phrase utilisateur, extrait les entités nommées (villes de départ et d'arrivée) via un modèle NER, puis calcule un itinéraire de transport optimal. Une option avancée permet de s'affranchir des API tierces en modélisant son propre graphe de connexions réseau ferroviaire.

Classification d'intentions NER Algorithmique de graphes
RECOMMENDER SYSTEMS SYSTÈMES DE RECOMMANDATION

RecFlix

Personal Perso Private

Movie recommendation engine utilizing cosine similarity and unsupervised learning via an API. Moteur de recommandation de films par similarité cosinus via une API asynchrone. Content-based movie recommendation system. The engine uses unsupervised learning to project and measure thematic and semantic similarities between films, served via a high-performance asynchronous API. Système de recommandation basé sur le contenu cinématographique (Content-Based Filtering). Le moteur utilise des méthodes d'apprentissage non-supervisé pour projeter et mesurer les proximités thématiques et sémantiques entre les œuvres, rendu disponible via une API asynchrone haute performance.

Unsupervised Learning Similarité Cosinus FastAPI
TEXT CLASSIFICATION / NLP TEXT CLASSIFICATION / NLP

Fake News Sniper

Personal Perso Private

NLP binary classifier evaluating linguistic and structural markers to detect fake news. Classifieur NLP de Fake News analysant les structures sémantiques et linguistiques. Semantic classification application trained to detect and isolate fake news. Evaluates text structures and suspicious linguistic markers to assign a credibility score to submitted articles. Application de classification sémantique entraînée pour identifier et isoler les fausses informations (Fake News). Le modèle évalue les structures textuelles et les marqueurs linguistiques suspects pour affecter un indice de fiabilité aux articles soumis.

NLP Classification Binaire Analyse Textuelle

Machine Learning & Tech Stack Stack Technique & Machine Learning

Deep Learning & Vision Deep Learning & Vision

  • PyTorch
    Astro-Retrieve WePlate MoodX
  • Deep Learning (CNN)
    MoodX Zoidberg 2.0
  • Computer Vision (OpenCV)
    MoodX Zoidberg 2.0
  • Algorithmes Génétiques
    Strategenetique

NLP & LLMs NLP & LLMs

  • Transformers & MLX
    Voltaire
  • LangChain & AI Agents
    Siquelator
  • Tokenisation & NER
    Voltaire Travel Order Resolver
  • Text-to-SQL
    Siquelator

MLOps & API Deployment MLOps & Déploiement

  • FastAPI
    Astro-Retrieve RecFlix
  • Ollama & Local LLMs
    Astro-Retrieve Siquelator
  • Pipelines MLOps
    Astro-Retrieve
  • Real-time Inference
    MoodX

Algorithms & Analysis Algorithmes & Analyse

  • Scikit-Learn
    Tune-Botrapoint
  • NumPy & Vectorization
    Strategenetique
  • Vector Databases
    Astro-Retrieve
  • Graph Algorithms
    Travel Order Resolver
  • Operations Research (OR)
    WePlate

Education & Credentials Formation & Titres

Academic Path Parcours Académique

MSc Pro in Artificial Intelligence MSc Pro en Intelligence Artificielle

Graduating Sept 2026 Diplôme prévu en Sept 2026

Epitech

Engineering Core Curriculum Tronc Commun Ingénieur

2021 — 2022

INSA Lyon (Electronics & Introduction to ML) INSA Lyon (Électronique & Introduction au ML)

CPGE (Math & Physics)

3-year intensive program Classe prépa intensive de 3 ans

Lycée Jean Perrin

Contact Me Contactez-moi

Let's build something together Construisons quelque chose ensemble

I am open to machine learning and software engineering opportunities. Whether you have a project idea, want to discuss LLM architectures, or just say hello, feel free to reach out! Je suis ouvert aux opportunités en machine learning et génie logiciel. Que vous ayez une idée de projet, souhaitiez discuter d'architectures de LLM ou simplement me saluer, n'hésitez pas à me contacter !

Location Localisation Montpellier / Toulon / Marseille / Nice / Remote