- Developing and implementing industry-specific data science solutions using GenAI and Machine Learning, ensuring seamless integration and deployment through robust MLOps practices.- Planning, conducting, and moderating client workshops, executing data science projects and proofs of concept, and presenting results to demonstrate the value of generative AI and machine learning solutions.- Identifying new business problems in the enterprise ML domain, collaborating with developers and product managers to create and standardize templates and frameworks from these innovative solutions.- Designing, developing, deploying, and maintaining advanced statistical and generative AI models, utilizing MLOps for continuous delivery, monitoring, and performance improvement.- Integrating IT ticket and document classification systems, implementing CI/CD pipelines, and keeping up-to-date with technological advancements in MLOps and generative AI to optimize enterprise applications.

RAG System Establishment: Development of a high-performance tool based on the RAG technique to help users quickly and easily find answers within a large volume of data (10GB).- Collection of metadata from various file types (docx, xlsx, html) for later use in our RAG model. - Development of two strategies for RAG: using Azure AI Search and PGVector. - Selection of the strategy based on Azure AI Search, achieving a correct response rate of 86.75%.

Classification of IT support tickets: Implementation of an IT ticket classification model and decision-making support.- Planning and moderating client workshops. - Establishing different strategies for ticket classification using various models (GPT-4, Embedding). - Analyzing results and detecting operator classification errors.- Achieved an accuracy of 73.2%, representing a 7.8% improvement over operator classification.

Documents’ Classification: Integration of a document analysis system for document classification and extraction of relevant information.- Processing insurance documents involving transformation, classification (˜100 classes), and extraction using GPT-Vision.- Accuracy on an evaluation system: 75% for classification and 90% for extraction. - Use of Databricks and MLFlow for deploying the model on a serving endpoint. - Creation and management of CI/CD pipelines on Azure DevOps to automate the deployment and update process.

Detection of accidents on construction sites: Creation and deployment of an ML model to predict potential accidents on various construction sites.- Integration and organization of client data for an improved database.- Data analysis, feature engineering, and adaptation of techniques for optimal performance. - Deployment, continuous improvement, and monitoring of models to detect any drift. - Implementation of automated re-training to maintain model performance.

Predictive maintenance: Optimize the replacement of grading tires to ensure continuous fleet availability.- Creation of data pipelines for client data transformation and conducting ML analyses and modeling. - Deployment of partitioned ML models with progressive improvements. - Monitoring of the model and ensuring its quality control. - Creation of a Tableau dashboard to provide the client with visual results.

- Developing industry specific data science solutions in enterprise ML. - Executing data science projects (showcases), PoCs (proof of concepts), and presenting results to Infor customers to demonstrate the value of the solutions (ML/optimization) in the enterprise ML domain. - Identifying and defining new business problems in the enterprise ML domain. - Collaborating with application developers, product managers on creating templates from these solutions. - Designing and developing statistical and mathematical models. - Deploying and maintaining the solutions. - Keeping up with advancements in technologies relevant to Infor and its enterprise ML customers. - Mentoring and guiding associate data scientists and interns. - Identifying internal and external data requirements and formulating data enrichment strategies for enterprise applications.

Modeling of P&L in the short term (3 seconds): Creation of a continuous delivery pipeline for real-time predictions of returns on the Chinese stock market.- Processing of high-frequency trading data for dynamic updates of the order book. - Identification of profitable trading scenarios & feature engineering. - Deployment of ML models with optimized scripts. - Implementation of automated checks to detect data and conceptual drifts.Implementation of a Python library: Profiling an internal Python library for daily data generation for the U.S. stock market.- Adding high-level functions to streamline data generation in more optimized scripts. - Improving memory usage and execution efficiency by approximately ~20%. - Validating the library's effectiveness by generating data objects daily.Analysis of the U.S. stock market:Analysis of trader behaviors on the U.S. stock market in response to large-scale ES contract transactions.- Identification of order types by tracking their historical evolution. - Estimation of the frequency of various order types occurring before and after large-scale ES contract transactions, with a 1 ms increment.Kalman Filter - Prediction of SSE50 volume: Supervision of an intern for replicating a research study using data from the SSE50 index of the Chinese stock market.- Thorough reading and understanding of the paper and identification of necessary data.- Breaking down the project into manageable steps. - Organizing weekly meetings to discuss progress and clarify doubts. - Encouraging multiple revisions of the work to refine results and improve the quality of the analysis.

- Creation and management of continuous delivery pipelines for real-time predictions.- Processing and updating trading data dynamically for financial markets.- Identifying and engineering features for profitable trading scenarios.- Deploying and optimizing machine learning models for financial applications.- Implementing automated systems for monitoring and detecting data anomalies and drifts.

- Worked on the project "Prediction of the Length of Stay in Hospitals."- Aimed to estimate the length of stay of patients using medical indicators.- Implemented several models, including Gradient Boosting, Random Forest, and Linear Regression.- Completed the project by creating a PowerBI solution to visualize and analyze the results.