Kevin Hua

- Audited and reworked a legacy opportunity spotting model, replacing rule-based establishment segmentation with clustering-based grouping and rebuilding spend benchmarking - Enriched and augmented data with AI Web scraping - Built and establish capture rate reporting across countries and operating models.

- Built an end-to-end deep learning workflow for large-scale multi-class image classification, benchmarking CNNs and Vision Transformers with MLflow tracking, a Streamlit app and Docker-based CI/CD on GitHub Actions. - Developed a spoken content understanding pipeline combining ASR, OCR, NLP and LLMs to turn audio and video into structured artefacts, including summarisation and RAG-style question answering over enriched transcripts. - Implemented classic ML projects on structured data (term-deposit classification, podcast listening time regression) covering EDA, feature engineering, cross-validated model selection and interpretation. - Created an interactive R Shiny showcasing clustering, regression, NLP, time series and probabilistic concepts through concise, reproducible dashboards.

• Forecasted sales by product and store using time series models (ARIMA, Prophet), improving inventory planning and stock allocation. • Designed and deployed pricing, net sales optimization and KPI monitoring models on GCP, leveraging BigQuery for large scale data processing and Vertex AI for model training and serving. • Collaborated with business stakeholders to translate analytical results into actionable strategies, define success metrics and support data-driven decision making.

• Partnered with Business teams to design and deliver an interactive NLP-based customer feedback analysis tool, providing granular insights and contributing to NPS improvement. • Developed and validated predictive models to optimize advertising campaigns, improving targeting efficiency and maximizing return on investment. • Built and implemented data quality monitoring models and controls for Big Data pipelines, ensuring reliability, consistency and auditability of key datasets.

• Developed machine learning models to detect fake news on Twitter, leveraging text mining and user-behaviour signals to improve classification performance. • Collaborated with the research team to design and maintain large-scale data collection and storage pipelines using MongoDB. • Benchmarked multiple ML algorithms, performed hyperparameter optimization (grid search), and improved model accuracy through feature engineering based on NLP, temporal patterns and regular expressions.

• Developed a flexible, interactive and adaptable R Shiny visualization tool to monitor the profitability of the insurance contracts. • Improved models predicting the probabilities of future claims by reducing prediction errors while respecting business constraints.

I was a part of an European project (Target 2 Securities) where I was asked to provide an analysis of the participants' behavior to highlight potential areas to improve the settlement of the engine. 4 major goals have been established and realized for my internship: • Provide reusable tools to analyze data from production • Analyze productions data to understand the use of T2S by the financial actors • Simulate new data from the analysis to test the settlement algorithms • Compare generated data sets with the statistic established profile

In a team of actuaries and data scientists, I participated in the conception of several modeling tools: • Study by Data-mining of an optical lenses database • Use of the Text-mining to classify disease treatments • Redaction of functional requirements for different softwares • Automation (under R) of data base's updates • Conception of a spraying tool in order to predict future charges