Mohamed KEITA

- Designed and implemented a scalable data infrastructure to move from local Dropbox-based storage to a centralized data lake in Google Cloud Storage (GCS). - Built an end-to-end ETL pipeline to ingest, clean, and transform heterogeneous datasets and store them efficiently in BigQuery as a data warehouse. - Deployed and fine-tuned a large language model (LLM) using Google Colab (with potential migration to GCP instances) to analyze CVs, job descriptions, and other sources to recommend tailored training paths for specific career goals. - Led the end-to-end MLOps lifecycle, including data engineering, model experimentation, and deployment workflows. - Managed production deployment and ensured infrastructure continuity within the Google Cloud Platform (GCP) ecosystem.

- Automated business processes to optimize group invoicing, invoice analysis, and the management of non-invoicing reasons, replacing manual operations (Excel formulas, VLOOKUPs) with efficient backend pipelines in PostgreSQL and SQL Server - Designed and maintained dynamic, interactive dashboards in Power BI, addressing key areas such as accounting, audit services, and patient data compliance, providing insights into key performance indicators - Managed data migration during the implementation of a new ERP, ensuring data quality and integrity throughout the transfer process - Authored detailed technical and functional documentation for all automated processes, enabling team members to easily understand and take over tasks when needed Administered databases: performed proactive maintenance, optimized performance, and resolved incidents to ensure the high availability of critical systems - Development of Two Chatbots: One for the Legal Department and One for the Patient Application - Data Collection, Cleaning, and Structuring (Retrieval of FAQs, technical documents) Fine-tuning of the LLM model, selection of an Azure-optimized LLM (Mistral AI) Integration of a RAG approach Deployment and monitoring on Azure Technical Environment: Python, Windows, PostgreSQL, SQL Server, PowerBI, Azure DevOps, Jira, Git

- Contribute in the creation of the back-end for the intranet using Django - Implement an ETL process to retrieve HR data and load it into a PostgreSQL database, which serves as the source for the intranet - Set up a CRON job to run the ETL process every morning - Create dashboards using PowerBI to display KPIs on the intranet - Data migration from a PostgreSQL database to Cloud SQL - Configure Google Cloud projects and services (Cloud SQL/BigQuery), ensuring robust security measures. - Adapt and transfer database schema and data using ETL tools - Perform comprehensive validation and performance testing to ensure data integrity and optimal performance. Technical Environment: Python, Django, SQL, Git, Windows, GCP, PostgreSQL, Scrum, PowerBI

- Conduct classes on MLOps workflow. - Mentore and guide students of the Data Science BootCamp in their projects. - Be a technical advisor, assisting students in developing machine learning models, putting them into production, and deploying them

Supervise a group of three second-year Master's students in Mathematical Engineering for their final year project. Main Tasks : - Guide students through the process and steps required for scientific research, helping them define objectives, problem statements, and formulating hypothesis. - Organize regular meetings based on the project's progress and the student’s needs. - Be a technical and theoretical advisor, overseeing students in developing mathematical models necessary for problem-solving and supporting them in writing a scientific report.

The project was aiming to identify inconsistencies among supplier invoices and sales invoices. Main tasks : - Create a pipeline to retrieve invoices from the email, manage billing errors, and store corrected lines in a MySQL database - Automating the pipeline to execute upon receiving new invoices in the email - Build a Python Flask graphical interface for manual error management by the accounting department Languages, Scripting & Frameworks : Python, Flask, SQL, Git Operating system & Tools : MacOS, AWS (S3) Database : MySQL, Cassandra

The project was aiming to building a matching application between travelers and hotels. Main Tasks : Build the matching algorithm to recommend a list of hotels to the user based on their traveler profile. - Build a weighted recommendation system. - Create an API (Routes and Endpoints) from the algorithms above - Conduct unit tests and continuous integration tests - Deploy continuously in Google Cloud - Evaluate the model continuously - Create a pipeline to retrieve hotel data from the back office (PostgreSQL database) and room availability from the Amadeus API - Store all this data in the MongoDB database, the source database for training various models - Ensure regular updates to the database with each change in room availability - Create dashboards with Tableau - Build new performance indicators (KPIs) - Index data in a PostgreSQL database Languages, Scripting & Frameworks : Python, Scikit-learn, Tensorflow, SQL, NoSQL, FastAPI, Git Operating system & Tools : Linux, GCP, Kubernetes Database : PostgreSQL, MongoDB

The project was aiming to give advices to the French government in order to design the future AI Master’s Main Tasks : - Explore the global patent database (PATSTAT) in a big data architecture with over 500 million observations Automating the pipeline to execute upon receiving new invoices in the email - Create a PySpark pipeline to retrieve data from the hard drive, perform all transformations necessary for identifying AI-related patents - Use the MLib library to apply models on data in Spark - Store identified AI patents in a SQLite database - Classify patents into different categories of artificial intelligence - Drive exploratory analysis to determine key technologies and players in the field of artificial intelligence - Publish a scientific report and a policy brief on artificial intelligence Languages, Scripting & Frameworks : Python, SQL, Spark, Hadoop Operating system & Tools : Linux, Windows Database : SQLite

Analysis of the results from the VICAN5 survey (life five years after a cancer diagnosis): Identifying patient profiles and observing their lifestyle habits. Identifying psychological factors that may lead to a relapse.