Régis Kla

Contexte & enjeux : Beaucoup d’entreprises ont un problème de données : peu fiables, doublonnées, mal stockées (beaucoup de perte), peu ou pas valorisées, etc. DIGITAL BRAIN souhaite fournir des services pour résoudre ces problèmes à l’aide de l’IA et de la Big Data. La marque Régis Kla Consulting propose des formations pointues qui permettent aux IT de se reconvertir dans les Métiers de la Donnée : AI Developer, Data Scientist, etc. Réalisations : - Co-création de la start-up DIGITAL BRAIN qui utilise l’IA et la Big Data à la résolution des problèmes de données des entreprises - Mise en place d’un framework d’Audit IT orienté données : Gouvernance IT, Gouvernance des Données, Plan Stratégique Informatique, etc - Prospection de partenariats, Business plans, Levée de fonds, etc - Développements de PoC(s) : - Système de Q&A ou chatbot par RAG et fine-tuning de petits LLMs, - Prototype d’une plateforme de Data as a Service (nettoyage à l’échelle, jointures 360 de données de différentes natures et sources, etc), - Plateforme de IA as a Service : APIs de KYC (Selfie verification, extraction de données à partir de CNI, etc) - Développements de formations à l’IA pour IT par la marque Régis Kla Consulting : - Becoming an AI Developer – Part I : From Zero to Hero / Part II : From Hero ro Superhero - An Introduction to Gen AI and LLM(s) – For Data Scientists and ML Engineers Env. Technique : GitOps (K8S), Cloud (GCP, Colab, Kaggle), LLMs (Ollama, Llama.*, Hugging Face, Weights & Biases), Big Data et Data lakes (Hadoop, HDFS), ML (TensorFlow, PyTorch), Go/Java/C++/Python...

Contexte & enjeux : DEVOLUTION est un pôle d’expertise en Ingénierie Logicielle (Développement et Intégration) au sein du Groupe NDA. Il intervient dans différents domaines Métiers dont la Banque, l’Assurance, la Santé, les NTIC… Le Groupe NDA se veut « orienté Données », en les plaçant au centre de sa stratégie de conception logicielle. Réalisations : Responsable de la Business Unit DEV&Innovation - Environ une vingtaine de personnes à gérer : Développeurs, Data Ingénieurs, Scrum masters, et une dizaine de projets en permanence. - Gestion de carrière : plans de formation continue, activités et objectifs, suivi des projets et management (COPIL, CODIR, etc), - Mise en place d’une Usine de Développement : Outils et processus GitOps, - Développement de PoCs IA : prédiction de flux de camions à un poste frontière. Env. Technique : Go/Java/C++, Cloud & GitOps (K8S, CI/CD), Hadoop, DXP, Web, Mobile, Microservices, ML (LLM, GPT4All, Llama) Env. Fonctionnel : Assurances, Technologie/Digital

Contexte & enjeux : * Le Datalab Group (DLG) développe des IA(s) pour l’ensemble du Groupe Crédit Agricole et ses Caisses. * Programme de certification des processus et IA(s) de DLG (https://www.lne.fr/fr/service/certification/certification-processus-ia). * L’IA n’est pas encore dans le « Business Model » de l’Entreprise Challenges, Difficultés : * Projet CACIB Search (délais courts, projet multi-départements, équipe multi-continent). * Pratiques avancées de GitOps et MLOps. Plusieurs IA(s) en production. Réalisations : * Création de pipelines de CI/CD pour projets ML en Python sur Gitlab EE. * MCO des IA(s) des clients. * Conception et développement d’une architecture orientée événements basée sur Kafka. * Développement de microservices pour exposer des IA(s) (contrôle automatique de CNI, Relevés d’Information d’assurance auto, de Cartes Grises…). * Support technique aux autres développeurs. Env. Technique : Gitlab CI, Cloud (K8s, Docker), Python, Poetry, MLflow, Kubeflow, Argo CD, GitOps (IaC) Env. Fonctionnel : Banques et Assurances

Department: EDF DOAAT/SIME/CSC-ASIE/DEVCO Achievements: * Development and maintenance of shared libraries in Python * Refactoring for larger tests coverage (JumOptim) : DDD and TDD * Creation of CI / CD pipelines for Python projects on Gitlab EE and Openshift * Technical support for Métier developers Technical Env: Gitlab EE, OpenShift, Python 3, PostgreSQL, AWS S3, MLflow, Airflow, SonarQube, Checkmarx, Hadoop, Spark Business Domain: Energy supply and distribution

NEO: e-Commerce Platform Context: * API-oriented e-Commerce platform in Production, to manage and sell services and insurance contracts (B2C, B2B2C) * The front ends are Web Sites, Mobile Applications, or CRM solutions * Project partners are working on the design of an operational and profitable model * 100% Agile and Scrum context (4 Dev teams), 100% Cloud Challenges, Difficulties: * 100% remote working mission, an international team: France, Spain, UK * Large project with a complex business * 100% continuous delivery (e.g. CI/CD) * Production context (BAU) and project (US and features) in parallel * Multidisciplinary team (Java, React, SalesForce) Achievements: * Full stack developer: Java, React (US and bugs fixes) * Tech lead (team velocity watch, team animation) Technical Env: Azure DevOps, AWS, OpenShift, SalesForce, PostgreSQL, Java 8 (Spring, Flyway, Gradle…), React, Docker Business Domain: Insurances and Services (Home & Motor, Consumer Electronics, Travel)

EDF DSIT DMA Solutions Lab Groupe Data Science IA (DSIA) Context: * DSIA contributes to the realization of AI projects for the business lines of all divisions and subsidiaries (framing, architecture, prototype development) * Projects types: Chat-bot, intelligent email processing, data processing (textual and images), and a (physical) robot to animate internal events and conferences at Picasso site * My role as Senior Data Scientist is to guarantee the technical and scientific realization of the projects: from the framing phase to the deployment in experimentation Challenges, Difficulties: * AI is not well known and is not yet fully accepted as a new "revolution", or simply is not yet in the (legacy) "Business Model" * The team is young and must master a new and difficult science (often empirical and intuitive); in a mature industrial context, extremely regulated, and codified Achievements: * C3PO Project: extending a legacy web application with a voice-based command system - Development of a command speech transcription system based on Deepspeech and ELMo, BERT - Deepspeech AI overtraining with Mozilla/Common-Voice-FR voices * E-BUY Project - Development of "Intelligent" search engines based on BoW, TF-IDF, ELMo, Word2Vec * Support/Training Junior Data Scientists and Python Developers Team: - Python, Machine and Deep Learning, Deployment of AI in Experimentation Technical Env: Linux, Docker, TensorFlow v1.x (GPU), Keras, Python (RAY, Flask, Gunicorn), Docker, Amazon Transcribe et Comprehend Business Domain: * Energy Supply

Sagemcom Energy & Telecom – URD Solutions - SICONIA/MDMS Context: The R&D Department developed the SICONIA Suite to provide additional server-side services for the Smart Metering Programme. The goal is to provide advanced analytics on the electricity, gas and water measurement data. And naturally the big data technologies are the right ones to handle such quantity of data (e.g. several Peta bytes on many projects). Challenges, Difficulties: * Resistance to change before SICONIA meters data were not considered as valuable * Data “preach” is required: the initial culture is not data oriented (e.g. it is hardware oriented) Achievements: * Design of a Lambda architecture based on Hadoop for SICONIA/MDM application - Batch and Serving layers design, sizing and prototyping for multi-size clients (from 4TB+ to 4PB+) - 2 Complete Hadoop DEV clusters installation and fine tuning: full-fledged or Docker containers based - Support for benchmarking strategy elaboration and design * Design of an automatic and continuous ML models training architecture for IoT - Collaboration with Data Scientists team - Use case: LoRa Geoloc or how to predict geographic position of IoT devices using LoRa protocol. - My role is to propose a scalable architecture to train automatically the prediction model(*) based on the everyday new coming data: 4TB+ for 100K devices every 6 days * Innovation internal Meet-ups: exchange and support around Deep Learning problematics (CNN, Face and Emotions Recognition) * Deep Learning prototype: Realization of a Speaker identification system based on CNN for Features extraction (Lite Cifar-10 NN) (*) A custom SVM used to detect outliers Technical Environment: Hadoop (TestDFSIO, NNBench), HBase, Flume, MapReduce (MRUnit), Spark/Java, Spark/Python, Spark-Bench, Solr, Jupyter, TensorFlow (Keras API), Deep Learning (CNN: LeNet, AlexNet, VGG16, Cifar-10) Functional Environment: Research n’ Development

BNP Paribas Securities Services – Digital Transformation Lab, Smart Data Context: As part of the Bank's Digital Transformation, a program for adding value to financial transactions data involving funds under management at the Bank is launched: SmartData. My role within the Data Research team is to support technically all PoCs launched by this program. The challenge is to guarantee their technical foundations. Challenges, Difficulties: * Access to data: resistance to change and data protection * Complexity of the Métiers related to the Finance: it is necessary to understand the business rules related to the data Achievements: * Creation and administration of the Data Factory - Installation of tools, and development processes definitions for Data Scientists in the Internal Cloud (Jupyterhub, Kernels R and Python, ELK, ...) - Definition of Big Data architecture patterns - Support to Data Scientist developments (code optimization, scaling) - System administration, packages installation, users management, Shell scripting and job submission (spark-submit) * Design and Deployment of "Data Product" in PoC mode: - Use case ISIN Brain (Predictive model of the daily value of a title): packaging in Web Service, Deployment - Use case ARC (Neural network for automatic recognition of graphics in PDFs): packaging as standalone Python application - Use Case WAVE (Web application for large-scale KPI aggregation): writing PySpark data preparation batchs, exposing business objects (DaaS: data as a service) on Elasticsearch, Web Services development, deployment * Collaboration with multidisciplinary teams (including the Business and IT), * Recruitment of Data Engineers Technical Environment: * Anaconda (Scikit-Learn, Pandas, Keras, OpenCV), R, Gitlab, Flask, Hadoop * MapR, Spark (Python, Scala, spark-submit), Jupyterhub, Elasticsearch, Apache HTTP Server, Flask, DRF Functional Environment: * Local and/or Global Custody, Settlement

CIB Digital Banking (BNPP/CIB ITO) Context: As part of the Digital Transformation of the CIB-ITO Department, the Digital Banking entity is created within the Project Management Department in order to monitor all initiatives related to Digital. The challenges are to study the viability of projects by conducting appropriate technical and functional studies. Challenges, Difficulties: * Structure of the Group: resistance to change, silotting of the professions and data, duplication of data and benchmarks * International dimension: developments in agile mode, and multi-country projects Achievements: * Solutions proposals, and Developments for Web and Mobile Applications (Centric Portal) - DB4C: Study of a single web portal for the Bank (multi-country portals, authentication, Id federation, delegation) - Fintech and Startups products integration solutions (BlueCircle) - Java development within Centric teams (Brussels, Chennai) in agile mode: Loan component - Supervision of a Java/J2ee developer trainee * Integration of Web Analytics solutions: interfaces, metrics (unique and return visitor conversion, bounce rate, ...) * Architecture Committees * Use Case TradeWeb (Predictive model to anticipate the outcome of a deal at a proposed price: win/lose classification) - DSSP graduation dissertation (*) - Business Background: Secondary market for government debts. Sales, purchases of Govies to institutional investors - Models: Logistic Regression, SVM, Decision Tree - Languages: Python, R (*) Confer Formations Technical Environment: Java/J2ee, HTML5/Angular 1.x, BPMN, EIP, SAML, OAuth2, Adobe Analytics, Google Analytics, Python, R Functional Environment: Investment Banking (Corporate Finance, Global Capital Markets, Structured Finance)

IRP Technical Architecture (BNPP/BP2S/AFS/IRP) Context: The IRP Information System (IS) is aging and is based on outdated technologies: Sybase, Transact-SQL, 4D. My mission as a technical architect is to propose a new architecture and the roadmap that takes into account reality: production constraints, team skills, outsourcing plans Challenges, Difficulties: * Organizational: business priorities vs. technical ones, an IT dependent of the business line * International dimension: offshore teams and follow-the-sun production Achievements: * IS tactical and strategic evolutions: Service-Oriented Architecture, ESB, BPM and BAM * Training, recruitment of local and offshore developers * Data integration, Prototyping, support for developments and production * Architecture committees Technical Environment: C/C++, Apache Camel, Sonar/Jenkins, BPMN, EIP, MongoDB, Jaspersoft, BIRT Functional Environment: Investment Banking, Custody, Settlement, Financial Reporting (VaR, PA, FIA,...)

Project MDSP (Orange, IBM) / Capgemini TMN Context: France Telecom - Orange as part of the redesign of its IS launches the MDSP project on all its 7 subsidiaries in Europe. The Migration team of which I am a part must migrate all the data of all the subscribers of all the subsidiaries without losing a single data Challenges, Difficulties: * Organizational: Deal won by IBM and Capgemini (Client's wish), often difficult coexistence * Technical: Management of "multi-all" (character sets, languages, encoding, ...) * Infrastructure: undersized for volume and migration windows expected by Customer Achievements: * Technical leader of the development team "Migration" * Choice of technologies and languages used * Migration of PIM data and mails (+ 7Tb) (multi-language, multi-country) * Setting up the platform and Continuous Integration processes Technical Environment: AIX, Perl, Shell, Maven 1.x, PMD, Checkstyle, CVS, PVCS, Java/J2ee, SLA High Performance Functional Environment: Telco: PIM (Perso Information Management) Data, SMS protocol

Internet racist and revisionist documents detection based on a Multi Agent System (MAS) - intelligent and autonomous agents design and implementation - design of agent based on syntactic and semantic analysis rules (INALCO) - design and realization of a CORBA 2.3 compliant MOM (simplified and fast JMS like protocol) Supervision of all developments (1 Ph. D. student, et 2 Postgraduate students) - support for one Ph.D. thesis prototype development in MAS and thesis document writing - support for postgraduate students prototype development and final work placement report Technical Environment : mod_jserv, Oracle 8i (PL/SQL), Debian Woody, Java, C/C++, KQML, CORBA 2.3, Threads

Seminars and practical works (DEUG MASS2) - OOP algorithmic using Java, JVM internals architecture (heap, garbage collection, etc), - abstract data type with Java (lists, trees, graphs, arrays) - Java GUI (AWT, Swing, Java2D) - design end of training project: develop a UML editor - students learns how to work in small teams to analyze requirements, design and develop - learn effective Swing (canvas, widget eraser, menus, event programming, ...) Technical Environment: Java, Sun Solaris, MS Windows